Datasets:

librarian-bots

/

dataset_abstracts

like 2

{ "abstract": "Improving the performance on an imbalanced training set is one of the main\nchallenges in nowadays Machine Learning. One way to augment and thus re-balance\nthe image dataset is through existing deep generative models, like\nclass-conditional Generative Adversarial Networks (cGAN) or Diffusion Models by\nsynthesizing images on each of the tail-class. Our experiments on imbalanced\nimage dataset classification show that, the validation accuracy improvement\nwith such re-balancing method is related to the image similarity between\ndifferent classes. Thus, to quantify this image dataset class similarity, we\npropose a measurement called Super-Sub Class Structural Similarity\n(SSIM-supSubCls) based on Structural Similarity (SSIM). A deep generative model\ndata augmentation classification (GM-augCls) pipeline is also provided to\nverify this metric correlates with the accuracy enhancement. We further\nquantify the relationship between them, discovering that the accuracy\nimprovement decays exponentially with respect to SSIM-supSubCls values.", "title": "Structural Similarity: When to Use Deep Generative Models on Imbalanced Image Dataset Augmentation", "url": "http://arxiv.org/abs/2303.04854v1" }

{ "text_length": 1172 }

{ "abstract": "Language has a profound impact on our thoughts, perceptions, and conceptions\nof gender roles. Gender-inclusive language is, therefore, a key tool to promote\nsocial inclusion and contribute to achieving gender equality. Consequently,\ndetecting and mitigating gender bias in texts is instrumental in halting its\npropagation and societal implications. However, there is a lack of gender bias\ndatasets and lexicons for automating the detection of gender bias using\nsupervised and unsupervised machine learning (ML) and natural language\nprocessing (NLP) techniques. Therefore, the main contribution of this work is\nto publicly provide labeled datasets and exhaustive lexicons by collecting,\nannotating, and augmenting relevant sentences to facilitate the detection of\ngender bias in English text. Towards this end, we present an updated version of\nour previously proposed taxonomy by re-formalizing its structure, adding a new\nbias type, and mapping each bias subtype to an appropriate detection\nmethodology. The released datasets and lexicons span multiple bias subtypes\nincluding: Generic He, Generic She, Explicit Marking of Sex, and Gendered\nNeologisms. We leveraged the use of word embedding models to further augment\nthe collected lexicons.", "title": "Gender Bias in Text: Labeled Datasets and Lexicons", "url": "http://arxiv.org/abs/2201.08675v2" }

{ "text_length": 1326 }

{ "abstract": "We introduce ArtBench-10, the first class-balanced, high-quality, cleanly\nannotated, and standardized dataset for benchmarking artwork generation. It\ncomprises 60,000 images of artwork from 10 distinctive artistic styles, with\n5,000 training images and 1,000 testing images per style. ArtBench-10 has\nseveral advantages over previous artwork datasets. Firstly, it is\nclass-balanced while most previous artwork datasets suffer from the long tail\nclass distributions. Secondly, the images are of high quality with clean\nannotations. Thirdly, ArtBench-10 is created with standardized data collection,\nannotation, filtering, and preprocessing procedures. We provide three versions\nof the dataset with different resolutions ($32\\times32$, $256\\times256$, and\noriginal image size), formatted in a way that is easy to be incorporated by\npopular machine learning frameworks. We also conduct extensive benchmarking\nexperiments using representative image synthesis models with ArtBench-10 and\npresent in-depth analysis. The dataset is available at\nhttps://github.com/liaopeiyuan/artbench under a Fair Use license.", "title": "The ArtBench Dataset: Benchmarking Generative Models with Artworks", "url": "http://arxiv.org/abs/2206.11404v1" }

{ "text_length": 1204 }

{ "abstract": "Machine learning applications cover a wide range of predictive tasks in which\ntabular datasets play a significant role. However, although they often address\nsimilar problems, tabular datasets are typically treated as standalone tasks.\nThe possibilities of using previously solved problems are limited due to the\nlack of structured contextual information about their features and the lack of\nunderstanding of the relations between them. To overcome this limitation, we\npropose a new approach called Semantic Feature Net (SeFNet), capturing the\nsemantic meaning of the analyzed tabular features. By leveraging existing\nontologies and domain knowledge, SeFNet opens up new opportunities for sharing\ninsights between diverse predictive tasks. One such opportunity is the Dataset\nOntology-based Semantic Similarity (DOSS) measure, which quantifies the\nsimilarity between datasets using relations across their features. In this\npaper, we present an example of SeFNet prepared for a collection of predictive\ntasks in healthcare, with the features' relations derived from the SNOMED-CT\nontology. The proposed SeFNet framework and the accompanying DOSS measure\naddress the issue of limited contextual information in tabular datasets. By\nincorporating domain knowledge and establishing semantic relations between\nfeatures, we enhance the potential for meta-learning and enable valuable\ninsights to be shared across different predictive tasks.", "title": "SeFNet: Bridging Tabular Datasets with Semantic Feature Nets", "url": "http://arxiv.org/abs/2306.11636v1" }

{ "text_length": 1527 }

{ "abstract": "Low light image enhancement is an important challenge for the development of\nrobust computer vision algorithms. The machine learning approaches to this have\nbeen either unsupervised, supervised based on paired dataset or supervised\nbased on unpaired dataset. This paper presents a novel deep learning pipeline\nthat can learn from both paired and unpaired datasets. Convolution Neural\nNetworks (CNNs) that are optimized to minimize standard loss, and Generative\nAdversarial Networks (GANs) that are optimized to minimize the adversarial loss\nare used to achieve different steps of the low light image enhancement process.\nCycle consistency loss and a patched discriminator are utilized to further\nimprove the performance. The paper also analyses the functionality and the\nperformance of different components, hidden layers, and the entire pipeline.", "title": "A Retinex based GAN Pipeline to Utilize Paired and Unpaired Datasets for Enhancing Low Light Images", "url": "http://arxiv.org/abs/2006.15304v2" }

{ "text_length": 981 }

{ "abstract": "As computers have become efficient at understanding visual information and\ntransforming it into a written representation, research interest in tasks like\nautomatic image captioning has seen a significant leap over the last few years.\nWhile most of the research attention is given to the English language in a\nmonolingual setting, resource-constrained languages like Bangla remain out of\nfocus, predominantly due to a lack of standard datasets. Addressing this issue,\nwe present a new dataset BAN-Cap following the widely used Flickr8k dataset,\nwhere we collect Bangla captions of the images provided by qualified\nannotators. Our dataset represents a wider variety of image caption styles\nannotated by trained people from different backgrounds. We present a\nquantitative and qualitative analysis of the dataset and the baseline\nevaluation of the recent models in Bangla image captioning. We investigate the\neffect of text augmentation and demonstrate that an adaptive attention-based\nmodel combined with text augmentation using Contextualized Word Replacement\n(CWR) outperforms all state-of-the-art models for Bangla image captioning. We\nalso present this dataset's multipurpose nature, especially on machine\ntranslation for Bangla-English and English-Bangla. This dataset and all the\nmodels will be useful for further research.", "title": "BAN-Cap: A Multi-Purpose English-Bangla Image Descriptions Dataset", "url": "http://arxiv.org/abs/2205.14462v1" }

{ "text_length": 1428 }

{ "abstract": "Multimodal deep learning systems which employ multiple modalities like text,\nimage, audio, video, etc., are showing better performance in comparison with\nindividual modalities (i.e., unimodal) systems. Multimodal machine learning\ninvolves multiple aspects: representation, translation, alignment, fusion, and\nco-learning. In the current state of multimodal machine learning, the\nassumptions are that all modalities are present, aligned, and noiseless during\ntraining and testing time. However, in real-world tasks, typically, it is\nobserved that one or more modalities are missing, noisy, lacking annotated\ndata, have unreliable labels, and are scarce in training or testing and or\nboth. This challenge is addressed by a learning paradigm called multimodal\nco-learning. The modeling of a (resource-poor) modality is aided by exploiting\nknowledge from another (resource-rich) modality using transfer of knowledge\nbetween modalities, including their representations and predictive models.\nCo-learning being an emerging area, there are no dedicated reviews explicitly\nfocusing on all challenges addressed by co-learning. To that end, in this work,\nwe provide a comprehensive survey on the emerging area of multimodal\nco-learning that has not been explored in its entirety yet. We review\nimplementations that overcome one or more co-learning challenges without\nexplicitly considering them as co-learning challenges. We present the\ncomprehensive taxonomy of multimodal co-learning based on the challenges\naddressed by co-learning and associated implementations. The various techniques\nemployed to include the latest ones are reviewed along with some of the\napplications and datasets. Our final goal is to discuss challenges and\nperspectives along with the important ideas and directions for future work that\nwe hope to be beneficial for the entire research community focusing on this\nexciting domain.", "title": "Multimodal Co-learning: Challenges, Applications with Datasets, Recent Advances and Future Directions", "url": "http://arxiv.org/abs/2107.13782v3" }

{ "text_length": 2031 }

{ "abstract": "Emotion recognition and understanding is a vital component in human-machine\ninteraction. Dimensional models of affect such as those using valence and\narousal have advantages over traditional categorical ones due to the complexity\nof emotional states in humans. However, dimensional emotion annotations are\ndifficult and expensive to collect, therefore they are not as prevalent in the\naffective computing community. To address these issues, we propose a method to\ngenerate synthetic images from existing categorical emotion datasets using face\nmorphing as well as dimensional labels in the circumplex space with full\ncontrol over the resulting sample distribution, while achieving augmentation\nfactors of at least 20x or more.", "title": "Morphset:Augmenting categorical emotion datasets with dimensional affect labels using face morphing", "url": "http://arxiv.org/abs/2103.02854v2" }

{ "text_length": 860 }

{ "abstract": "While the whole world is still struggling with the COVID-19 pandemic, online\nlearning and home office become more common. Many schools transfer their\ncourses teaching to the online classroom. Therefore, it is significant to mine\nthe students' feedback and opinions from their reviews towards studies so that\nboth schools and teachers can know where they need to improve. This paper\ntrains machine learning and deep learning models using both balanced and\nimbalanced datasets for sentiment classification. Two SOTA category-aware text\ngeneration GAN models: CatGAN and SentiGAN, are utilized to synthesize text\nused to balance the highly imbalanced dataset. Results on three datasets with\ndifferent imbalance degree from distinct domains show that when using generated\ntext to balance the dataset, the F1-score of machine learning and deep learning\nmodel on sentiment classification increases 2.79% ~ 9.28%. Also, the results\nindicate that the average growth degree for CR100k is higher than CR23k, the\naverage growth degree for deep learning is more increased than machine learning\nalgorithms, and the average growth degree for more complex deep learning models\nis more increased than simpler deep learning models in experiments.", "title": "Using GAN-based models to sentimental analysis on imbalanced datasets in education domain", "url": "http://arxiv.org/abs/2108.12061v1" }

{ "text_length": 1353 }

{ "abstract": "Many physics-based and machine-learned scoring functions (SFs) used to\npredict protein-ligand binding free energies have been trained on the PDBBind\ndataset. However, it is controversial as to whether new SFs are actually\nimproving since the general, refined, and core datasets of PDBBind are\ncross-contaminated with proteins and ligands with high similarity, and hence\nthey may not perform comparably well in binding prediction of new\nprotein-ligand complexes. In this work we have carefully prepared a cleaned\nPDBBind data set of non-covalent binders that are split into training,\nvalidation, and test datasets to control for data leakage. The resulting\nleak-proof (LP)-PDBBind data is used to retrain four popular SFs: AutoDock\nvina, Random Forest (RF)-Score, InteractionGraphNet (IGN), and DeepDTA, to\nbetter test their capabilities when applied to new protein-ligand complexes. In\nparticular we have formulated a new independent data set, BDB2020+, by matching\nhigh quality binding free energies from BindingDB with co-crystalized\nligand-protein complexes from the PDB that have been deposited since 2020.\nBased on all the benchmark results, the retrained models using LP-PDBBind that\nrely on 3D information perform consistently among the best, with IGN especially\nbeing recommended for scoring and ranking applications for new protein-ligand\nsystems.", "title": "Leak Proof PDBBind: A Reorganized Dataset of Protein-Ligand Complexes for More Generalizable Binding Affinity Prediction", "url": "http://arxiv.org/abs/2308.09639v1" }

{ "text_length": 1511 }

{ "abstract": "In the last decade, machine learning-based approaches became capable of\nperforming a wide range of complex tasks sometimes better than humans,\ndemanding a fraction of the time. Such an advance is partially due to the\nexponential growth in the amount of data available, which makes it possible to\nextract trustworthy real-world information from them. However, such data is\ngenerally imbalanced since some phenomena are more likely than others. Such a\nbehavior yields considerable influence on the machine learning model's\nperformance since it becomes biased on the more frequent data it receives.\nDespite the considerable amount of machine learning methods, a graph-based\napproach has attracted considerable notoriety due to the outstanding\nperformance over many applications, i.e., the Optimum-Path Forest (OPF). In\nthis paper, we propose three OPF-based strategies to deal with the imbalance\nproblem: the $\\text{O}^2$PF and the OPF-US, which are novel approaches for\noversampling and undersampling, respectively, as well as a hybrid strategy\ncombining both approaches. The paper also introduces a set of variants\nconcerning the strategies mentioned above. Results compared against several\nstate-of-the-art techniques over public and private datasets confirm the\nrobustness of the proposed approaches.", "title": "Handling Imbalanced Datasets Through Optimum-Path Forest", "url": "http://arxiv.org/abs/2202.08934v1" }

{ "text_length": 1392 }

{ "abstract": "This paper presents HaVQA, the first multimodal dataset for visual\nquestion-answering (VQA) tasks in the Hausa language. The dataset was created\nby manually translating 6,022 English question-answer pairs, which are\nassociated with 1,555 unique images from the Visual Genome dataset. As a\nresult, the dataset provides 12,044 gold standard English-Hausa parallel\nsentences that were translated in a fashion that guarantees their semantic\nmatch with the corresponding visual information. We conducted several baseline\nexperiments on the dataset, including visual question answering, visual\nquestion elicitation, text-only and multimodal machine translation.", "title": "HaVQA: A Dataset for Visual Question Answering and Multimodal Research in Hausa Language", "url": "http://arxiv.org/abs/2305.17690v1" }

{ "text_length": 778 }

{ "abstract": "Automated searches for strong gravitational lensing in optical imaging survey\ndatasets often employ machine learning and deep learning approaches. These\ntechniques require more example systems to train the algorithms than have\npresently been discovered, which creates a need for simulated images as\ntraining dataset supplements. This work introduces and summarizes\ndeeplenstronomy, an open-source Python package that enables efficient,\nlarge-scale, and reproducible simulation of images of astronomical systems. A\nfull suite of unit tests, documentation, and example notebooks are available at\nhttps://deepskies.github.io/deeplenstronomy/ .", "title": "deeplenstronomy: A dataset simulation package for strong gravitational lensing", "url": "http://arxiv.org/abs/2102.02830v1" }

{ "text_length": 753 }

{ "abstract": "Machine Learning (ML) is accelerating progress across fields and industries,\nbut relies on accessible and high-quality training data. Some of the most\nimportant datasets are found in biomedical and financial domains in the form of\nspreadsheets and relational databases. But this tabular data is often sensitive\nin nature. Synthetic data generation offers the potential to unlock sensitive\ndata, but generative models tend to memorise and regurgitate training data,\nwhich undermines the privacy goal. To remedy this, researchers have\nincorporated the mathematical framework of Differential Privacy (DP) into the\ntraining process of deep neural networks. But this creates a trade-off between\nthe quality and privacy of the resulting data. Generative Adversarial Networks\n(GANs) are the dominant paradigm for synthesising tabular data under DP, but\nsuffer from unstable adversarial training and mode collapse, which are\nexacerbated by the privacy constraints and challenging tabular data modality.\nThis work optimises the quality-privacy trade-off of generative models,\nproducing higher quality tabular datasets with the same privacy guarantees. We\nimplement novel end-to-end models that leverage attention mechanisms to learn\nreversible tabular representations. We also introduce TableDiffusion, the first\ndifferentially-private diffusion model for tabular data synthesis. Our\nexperiments show that TableDiffusion produces higher-fidelity synthetic\ndatasets, avoids the mode collapse problem, and achieves state-of-the-art\nperformance on privatised tabular data synthesis. By implementing\nTableDiffusion to predict the added noise, we enabled it to bypass the\nchallenges of reconstructing mixed-type tabular data. Overall, the diffusion\nparadigm proves vastly more data and privacy efficient than the adversarial\nparadigm, due to augmented re-use of each data batch and a smoother iterative\ntraining process.", "title": "Generating tabular datasets under differential privacy", "url": "http://arxiv.org/abs/2308.14784v1" }

{ "text_length": 1995 }

{ "abstract": "This dataset contains anonymised human lung computed tomography (CT) scans\nwith COVID-19 related findings, as well as without such findings. A small\nsubset of studies has been annotated with binary pixel masks depicting regions\nof interests (ground-glass opacifications and consolidations). CT scans were\nobtained between 1st of March, 2020 and 25th of April, 2020, and provided by\nmunicipal hospitals in Moscow, Russia. Permanent link:\nhttps://mosmed.ai/datasets/covid19_1110. This dataset is licensed under a\nCreative Commons Attribution-NonCommercial-NoDerivs 3.0 Unported (CC BY-NC-ND\n3.0) License. Key words: artificial intelligence, COVID-19, machine learning,\ndataset, CT, chest, imaging", "title": "MosMedData: Chest CT Scans With COVID-19 Related Findings Dataset", "url": "http://arxiv.org/abs/2005.06465v1" }

{ "text_length": 794 }

{ "abstract": "Surrogate models are necessary to optimize meaningful quantities in physical\ndynamics as their recursive numerical resolutions are often prohibitively\nexpensive. It is mainly the case for fluid dynamics and the resolution of\nNavier-Stokes equations. However, despite the fast-growing field of data-driven\nmodels for physical systems, reference datasets representing real-world\nphenomena are lacking. In this work, we develop AirfRANS, a dataset for\nstudying the two-dimensional incompressible steady-state Reynolds-Averaged\nNavier-Stokes equations over airfoils at a subsonic regime and for different\nangles of attacks. We also introduce metrics on the stress forces at the\nsurface of geometries and visualization of boundary layers to assess the\ncapabilities of models to accurately predict the meaningful information of the\nproblem. Finally, we propose deep learning baselines on four machine learning\ntasks to study AirfRANS under different constraints for generalization\nconsiderations: big and scarce data regime, Reynolds number, and angle of\nattack extrapolation.", "title": "AirfRANS: High Fidelity Computational Fluid Dynamics Dataset for Approximating Reynolds-Averaged Navier-Stokes Solutions", "url": "http://arxiv.org/abs/2212.07564v3" }

{ "text_length": 1225 }

{ "abstract": "In this paper, we consider the intersection of two problems in machine\nlearning: Multi-Source Domain Adaptation (MSDA) and Dataset Distillation (DD).\nOn the one hand, the first considers adapting multiple heterogeneous labeled\nsource domains to an unlabeled target domain. On the other hand, the second\nattacks the problem of synthesizing a small summary containing all the\ninformation about the datasets. We thus consider a new problem called MSDA-DD.\nTo solve it, we adapt previous works in the MSDA literature, such as\nWasserstein Barycenter Transport and Dataset Dictionary Learning, as well as DD\nmethod Distribution Matching. We thoroughly experiment with this novel problem\non four benchmarks (Caltech-Office 10, Tennessee-Eastman Process, Continuous\nStirred Tank Reactor, and Case Western Reserve University), where we show that,\neven with as little as 1 sample per class, one achieves state-of-the-art\nadaptation performance.", "title": "Multi-Source Domain Adaptation meets Dataset Distillation through Dataset Dictionary Learning", "url": "http://arxiv.org/abs/2309.07666v1" }

{ "text_length": 1062 }

{ "abstract": "Fighting online hate speech is a challenge that is usually addressed using\nNatural Language Processing via automatic detection and removal of hate\ncontent. Besides this approach, counter narratives have emerged as an effective\ntool employed by NGOs to respond to online hate on social media platforms. For\nthis reason, Natural Language Generation is currently being studied as a way to\nautomatize counter narrative writing. However, the existing resources necessary\nto train NLG models are limited to 2-turn interactions (a hate speech and a\ncounter narrative as response), while in real life, interactions can consist of\nmultiple turns. In this paper, we present a hybrid approach for dialogical data\ncollection, which combines the intervention of human expert annotators over\nmachine generated dialogues obtained using 19 different configurations. The\nresult of this work is DIALOCONAN, the first dataset comprising over 3000\nfictitious multi-turn dialogues between a hater and an NGO operator, covering 6\ntargets of hate.", "title": "Human-Machine Collaboration Approaches to Build a Dialogue Dataset for Hate Speech Countering", "url": "http://arxiv.org/abs/2211.03433v1" }

{ "text_length": 1152 }

{ "abstract": "Effective environmental planning and management to address climate change\ncould be achieved through extensive environmental modeling with machine\nlearning and conventional physical models. In order to develop and improve\nthese models, practitioners and researchers need comprehensive benchmark\ndatasets that are prepared and processed with environmental expertise that they\ncan rely on. This study presents an extensive dataset of rainfall events for\nthe state of Iowa (2016-2019) acquired from the National Weather Service Next\nGeneration Weather Radar (NEXRAD) system and processed by a quantitative\nprecipitation estimation system. The dataset presented in this study could be\nused for better disaster monitoring, response and recovery by paving the way\nfor both predictive and prescriptive modeling.", "title": "IowaRain: A Statewide Rain Event Dataset Based on Weather Radars and Quantitative Precipitation Estimation", "url": "http://arxiv.org/abs/2107.03432v1" }

{ "text_length": 944 }

{ "abstract": "As machine learning methods gain prominence within clinical decision-making,\naddressing fairness concerns becomes increasingly urgent. Despite considerable\nwork dedicated to detecting and ameliorating algorithmic bias, today's methods\nare deficient with potentially harmful consequences. Our causal perspective\nsheds new light on algorithmic bias, highlighting how different sources of\ndataset bias may appear indistinguishable yet require substantially different\nmitigation strategies. We introduce three families of causal bias mechanisms\nstemming from disparities in prevalence, presentation, and annotation. Our\ncausal analysis underscores how current mitigation methods tackle only a narrow\nand often unrealistic subset of scenarios. We provide a practical three-step\nframework for reasoning about fairness in medical imaging, supporting the\ndevelopment of safe and equitable AI prediction models.", "title": "No Fair Lunch: A Causal Perspective on Dataset Bias in Machine Learning for Medical Imaging", "url": "http://arxiv.org/abs/2307.16526v1" }

{ "text_length": 1028 }

{ "abstract": "Rapid advancement of machine learning solutions has often coincided with the\nproduction of a test public data set. Such datasets reduce the largest barrier\nto entry for tackling a problem -- procuring data -- while also providing a\nbenchmark to compare different solutions. Furthermore, large datasets have been\nused to train high-performing feature finders which are then used in new\napproaches to problems beyond that initially defined. In order to encourage the\nrapid development in the analysis of data collected using liquid argon time\nprojection chambers, a class of particle detectors used in high energy physics\nexperiments, we have produced the PILArNet, first 2D and 3D open dataset to be\nused for a couple of key analysis tasks. The initial dataset presented in this\npaper contains 300,000 samples simulated and recorded in three different volume\nsizes. The dataset is stored efficiently in sparse 2D and 3D matrix format with\nauxiliary information about simulated particles in the volume, and is made\navailable for public research use. In this paper we describe the dataset,\ntasks, and the method used to procure the sample.", "title": "PILArNet: Public Dataset for Particle Imaging Liquid Argon Detectors in High Energy Physics", "url": "http://arxiv.org/abs/2006.01993v1" }

{ "text_length": 1262 }

{ "abstract": "The need for secure Internet of Things (IoT) devices is growing as IoT\ndevices are becoming more integrated into vital networks. Many systems rely on\nthese devices to remain available and provide reliable service. Denial of\nservice attacks against IoT devices are a real threat due to the fact these low\npower devices are very susceptible to denial-of-service attacks. Machine\nlearning enabled network intrusion detection systems are effective at\nidentifying new threats, but they require a large amount of data to work well.\nThere are many network traffic data sets but very few that focus on IoT network\ntraffic. Within the IoT network data sets there is a lack of CoAP denial of\nservice data. We propose a novel data set covering this gap. We develop a new\ndata set by collecting network traffic from real CoAP denial of service attacks\nand compare the data on multiple different machine learning classifiers. We\nshow that the data set is effective on many classifiers.", "title": "CoAP-DoS: An IoT Network Intrusion Dataset", "url": "http://arxiv.org/abs/2206.14341v1" }

{ "text_length": 1049 }

{ "abstract": "Nowadays, many researchers are focusing their attention on the subject of\nmachine translation (MT). However, Persian machine translation has remained\nunexplored despite a vast amount of research being conducted in languages with\nhigh resources, such as English. Moreover, while a substantial amount of\nresearch has been undertaken in statistical machine translation for some\ndatasets in Persian, there is currently no standard baseline for\ntransformer-based text2text models on each corpus. This study collected and\nanalysed the most popular and valuable parallel corpora, which were used for\nPersian-English translation. Furthermore, we fine-tuned and evaluated two\nstate-of-the-art attention-based seq2seq models on each dataset separately (48\nresults). We hope this paper will assist researchers in comparing their Persian\nto English and vice versa machine translation results to a standard baseline.", "title": "An Evaluation of Persian-English Machine Translation Datasets with Transformers", "url": "http://arxiv.org/abs/2302.00321v1" }

{ "text_length": 1017 }

{ "abstract": "Data is commonly stored in tabular format. Several fields of research are\nprone to small imbalanced tabular data. Supervised Machine Learning on such\ndata is often difficult due to class imbalance. Synthetic data generation,\ni.e., oversampling, is a common remedy used to improve classifier performance.\nState-of-the-art linear interpolation approaches, such as LoRAS and ProWRAS can\nbe used to generate synthetic samples from the convex space of the minority\nclass to improve classifier performance in such cases. Deep generative networks\nare common deep learning approaches for synthetic sample generation, widely\nused for synthetic image generation. However, their scope on synthetic tabular\ndata generation in the context of imbalanced classification is not adequately\nexplored. In this article, we show that existing deep generative models perform\npoorly compared to linear interpolation based approaches for imbalanced\nclassification problems on smaller tabular datasets. To overcome this, we\npropose a deep generative model, ConvGeN that combines the idea of convex space\nlearning with deep generative models. ConvGeN learns the coefficients for the\nconvex combinations of the minority class samples, such that the synthetic data\nis distinct enough from the majority class. Our benchmarking experiments\ndemonstrate that our proposed model ConvGeN improves imbalanced classification\non such small datasets, as compared to existing deep generative models, while\nbeing at-par with the existing linear interpolation approaches. Moreover, we\ndiscuss how our model can be used for synthetic tabular data generation in\ngeneral, even outside the scope of data imbalance and thus, improves the\noverall applicability of convex space learning.", "title": "ConvGeN: Convex space learning improves deep-generative oversampling for tabular imbalanced classification on smaller datasets", "url": "http://arxiv.org/abs/2206.09812v2" }

{ "text_length": 1900 }

{ "abstract": "Motivation: Novel machine learning and statistical modeling studies rely on\nstandardized comparisons to existing methods using well-studied benchmark\ndatasets. Few tools exist that provide rapid access to many of these datasets\nthrough a standardized, user-friendly interface that integrates well with\npopular data science workflows.\n Results: This release of PMLB provides the largest collection of diverse,\npublic benchmark datasets for evaluating new machine learning and data science\nmethods aggregated in one location. v1.0 introduces a number of critical\nimprovements developed following discussions with the open-source community.\n Availability: PMLB is available at https://github.com/EpistasisLab/pmlb.\nPython and R interfaces for PMLB can be installed through the Python Package\nIndex and Comprehensive R Archive Network, respectively.", "title": "PMLB v1.0: An open source dataset collection for benchmarking machine learning methods", "url": "http://arxiv.org/abs/2012.00058v3" }

{ "text_length": 968 }

{ "abstract": "In the last 150 years, CO2 concentration in the atmosphere has increased from\n280 parts per million to 400 parts per million. This has caused an increase in\nthe average global temperatures by nearly 0.7 degree centigrade due to the\ngreenhouse effect. However, the most prosperous states are the highest emitters\nof greenhouse gases (specially, CO2). This indicates a strong relationship\nbetween gaseous emissions and the gross domestic product (GDP) of the states.\nSuch a relationship is highly volatile and nonlinear due to its dependence on\nthe technological advancements and constantly changing domestic and\ninternational regulatory policies and relations. To analyse such vastly\nnonlinear relationships, soft computing techniques has been quite effective as\nthey can predict a compact solution for multi-variable parameters without any\nexplicit insight into the internal system functionalities. This paper reports a\nnovel transfer learning based approach for GDP prediction, which we have termed\nas Domain Adapted Transfer Learning for GDP Prediction. In the proposed\napproach per capita GDP of different nations is predicted using their CO2\nemissions via a model trained on the data of any developed or developing\neconomy. Results are comparatively presented considering three well-known\nregression methods such as Generalized Regression Neural Network, Extreme\nLearning Machine and Support Vector Regression. Then the proposed approach is\nused to reliably estimate the missing per capita GDP of some of the war-torn\nand isolated countries.", "title": "A Novel GDP Prediction Technique based on Transfer Learning using CO2 Emission Dataset", "url": "http://arxiv.org/abs/2005.02856v1" }

{ "text_length": 1666 }

{ "abstract": "Mobile machines typically working in a closed site, have a high potential to\nutilize autonomous driving technology. However, vigorously thriving development\nand innovation are happening mostly in the area of passenger cars. In contrast,\nalthough there are also many research pieces about autonomous driving or\nworking in mobile machines, a consensus about the SOTA solution is still not\nachieved. We believe that the most urgent problem that should be solved is the\nabsence of a public and challenging visual dataset, which makes the results\nfrom different researches comparable. To address the problem, we publish the\nKIT MOMA dataset, including eight classes of commonly used mobile machines,\nwhich can be used as a benchmark to evaluate the SOTA algorithms to detect\nmobile construction machines. The view of the gathered images is outside of the\nmobile machines since we believe fixed cameras on the ground are more suitable\nif all the interesting machines are working in a closed site. Most of the\nimages in KIT MOMA are in a real scene, whereas some of the images are from the\nofficial website of top construction machine companies. Also, we have evaluated\nthe performance of YOLO v3 on our dataset, indicating that the SOTA computer\nvision algorithms already show an excellent performance for detecting the\nmobile machines in a specific working site. Together with the dataset, we also\nupload the trained weights, which can be directly used by engineers from the\nconstruction machine industry. The dataset, trained weights, and updates can be\nfound on our Github. Moreover, the demo can be found on our Youtube.", "title": "KIT MOMA: A Mobile Machines Dataset", "url": "http://arxiv.org/abs/2007.04198v1" }

{ "text_length": 1688 }

{ "abstract": "Recurrent graph convolutional neural networks are highly effective machine\nlearning techniques for spatiotemporal signal processing. Newly proposed graph\nneural network architectures are repetitively evaluated on standard tasks such\nas traffic or weather forecasting. In this paper, we propose the Chickenpox\nCases in Hungary dataset as a new dataset for comparing graph neural network\narchitectures. Our time series analysis and forecasting experiments demonstrate\nthat the Chickenpox Cases in Hungary dataset is adequate for comparing the\npredictive performance and forecasting capabilities of novel recurrent graph\nneural network architectures.", "title": "Chickenpox Cases in Hungary: a Benchmark Dataset for Spatiotemporal Signal Processing with Graph Neural Networks", "url": "http://arxiv.org/abs/2102.08100v1" }

{ "text_length": 794 }

{ "abstract": "We study scalable machine learning models for full event reconstruction in\nhigh-energy electron-positron collisions based on a highly granular detector\nsimulation. Particle-flow (PF) reconstruction can be formulated as a supervised\nlearning task using tracks and calorimeter clusters or hits. We compare a graph\nneural network and kernel-based transformer and demonstrate that both avoid\nquadratic memory allocation and computational cost while achieving realistic PF\nreconstruction. We show that hyperparameter tuning on a supercomputer\nsignificantly improves the physics performance of the models. We also\ndemonstrate that the resulting model is highly portable across hardware\nprocessors, supporting Nvidia, AMD, and Intel Habana cards. Finally, we\ndemonstrate that the model can be trained on highly granular inputs consisting\nof tracks and calorimeter hits, resulting in a competitive physics performance\nwith the baseline. Datasets and software to reproduce the studies are published\nfollowing the findable, accessible, interoperable, and reusable (FAIR)\nprinciples.", "title": "Scalable neural network models and terascale datasets for particle-flow reconstruction", "url": "http://arxiv.org/abs/2309.06782v1" }

{ "text_length": 1193 }

{ "abstract": "Legal judgment prediction (LJP) applies Natural Language Processing (NLP)\ntechniques to predict judgment results based on fact descriptions\nautomatically. Recently, large-scale public datasets and advances in NLP\nresearch have led to increasing interest in LJP. Despite a clear gap between\nmachine and human performance, impressive results have been achieved in various\nbenchmark datasets. In this paper, to address the current lack of comprehensive\nsurvey of existing LJP tasks, datasets, models and evaluations, (1) we analyze\n31 LJP datasets in 6 languages, present their construction process and define a\nclassification method of LJP with 3 different attributes; (2) we summarize 14\nevaluation metrics under four categories for different outputs of LJP tasks;\n(3) we review 12 legal-domain pretrained models in 3 languages and highlight 3\nmajor research directions for LJP; (4) we show the state-of-art results for 8\nrepresentative datasets from different court cases and discuss the open\nchallenges. This paper can provide up-to-date and comprehensive reviews to help\nreaders understand the status of LJP. We hope to facilitate both NLP\nresearchers and legal professionals for further joint efforts in this problem.", "title": "A Survey on Legal Judgment Prediction: Datasets, Metrics, Models and Challenges", "url": "http://arxiv.org/abs/2204.04859v1" }

{ "text_length": 1334 }

{ "abstract": "As free online encyclopedias with massive volumes of content, Wikipedia and\nWikidata are key to many Natural Language Processing (NLP) tasks, such as\ninformation retrieval, knowledge base building, machine translation, text\nclassification, and text summarization. In this paper, we introduce WikiDes, a\nnovel dataset to generate short descriptions of Wikipedia articles for the\nproblem of text summarization. The dataset consists of over 80k English samples\non 6987 topics. We set up a two-phase summarization method - description\ngeneration (Phase I) and candidate ranking (Phase II) - as a strong approach\nthat relies on transfer and contrastive learning. For description generation,\nT5 and BART show their superiority compared to other small-scale pre-trained\nmodels. By applying contrastive learning with the diverse input from beam\nsearch, the metric fusion-based ranking models outperform the direct\ndescription generation models significantly up to 22 ROUGE in topic-exclusive\nsplit and topic-independent split. Furthermore, the outcome descriptions in\nPhase II are supported by human evaluation in over 45.33% chosen compared to\n23.66% in Phase I against the gold descriptions. In the aspect of sentiment\nanalysis, the generated descriptions cannot effectively capture all sentiment\npolarities from paragraphs while doing this task better from the gold\ndescriptions. The automatic generation of new descriptions reduces the human\nefforts in creating them and enriches Wikidata-based knowledge graphs. Our\npaper shows a practical impact on Wikipedia and Wikidata since there are\nthousands of missing descriptions. Finally, we expect WikiDes to be a useful\ndataset for related works in capturing salient information from short\nparagraphs. The curated dataset is publicly available at:\nhttps://github.com/declare-lab/WikiDes.", "title": "WikiDes: A Wikipedia-Based Dataset for Generating Short Descriptions from Paragraphs", "url": "http://arxiv.org/abs/2209.13101v1" }

{ "text_length": 1949 }

{ "abstract": "The number of IoT devices in healthcare is expected to rise sharply due to\nincreased demand since the COVID-19 pandemic. Deep learning and IoT devices are\nbeing employed to monitor body vitals and automate anomaly detection in\nclinical and non-clinical settings. Most of the current technology requires the\ntransmission of raw data to a remote server, which is not efficient for\nresource-constrained IoT devices and embedded systems. Additionally, it is\nchallenging to develop a machine learning model for ECG classification due to\nthe lack of an extensive open public database. To an extent, to overcome this\nchallenge PTB-XL dataset has been used. In this work, we have developed machine\nlearning models to be deployed on Raspberry Pi. We present an evaluation of our\nTensorFlow Model with two classification classes. We also present the\nevaluation of the corresponding TensorFlow Lite FlatBuffers to demonstrate\ntheir minimal run-time requirements while maintaining acceptable accuracy.", "title": "Deep Learning-based ECG Classification on Raspberry PI using a Tensorflow Lite Model based on PTB-XL Dataset", "url": "http://arxiv.org/abs/2209.00989v1" }

{ "text_length": 1132 }

{ "abstract": "This paper presents a new training dataset for automatic genre identification\nGINCO, which is based on 1,125 crawled Slovenian web documents that consist of\n650 thousand words. Each document was manually annotated for genre with a new\nannotation schema that builds upon existing schemata, having primarily clarity\nof labels and inter-annotator agreement in mind. The dataset consists of\nvarious challenges related to web-based data, such as machine translated\ncontent, encoding errors, multiple contents presented in one document etc.,\nenabling evaluation of classifiers in realistic conditions. The initial machine\nlearning experiments on the dataset show that (1) pre-Transformer models are\ndrastically less able to model the phenomena, with macro F1 metrics ranging\naround 0.22, while Transformer-based models achieve scores of around 0.58, and\n(2) multilingual Transformer models work as well on the task as the monolingual\nmodels that were previously proven to be superior to multilingual models on\nstandard NLP tasks.", "title": "The GINCO Training Dataset for Web Genre Identification of Documents Out in the Wild", "url": "http://arxiv.org/abs/2201.03857v1" }

{ "text_length": 1142 }

{ "abstract": "Machine learning is predicated on the concept of generalization: a model\nachieving low error on a sufficiently large training set should also perform\nwell on novel samples from the same distribution. We show that both data\nwhitening and second order optimization can harm or entirely prevent\ngeneralization. In general, model training harnesses information contained in\nthe sample-sample second moment matrix of a dataset. For a general class of\nmodels, namely models with a fully connected first layer, we prove that the\ninformation contained in this matrix is the only information which can be used\nto generalize. Models trained using whitened data, or with certain second order\noptimization schemes, have less access to this information, resulting in\nreduced or nonexistent generalization ability. We experimentally verify these\npredictions for several architectures, and further demonstrate that\ngeneralization continues to be harmed even when theoretical requirements are\nrelaxed. However, we also show experimentally that regularized second order\noptimization can provide a practical tradeoff, where training is accelerated\nbut less information is lost, and generalization can in some circumstances even\nimprove.", "title": "Whitening and second order optimization both make information in the dataset unusable during training, and can reduce or prevent generalization", "url": "http://arxiv.org/abs/2008.07545v4" }

{ "text_length": 1396 }

{ "abstract": "Code is increasingly becoming a core data modality of modern machine learning\nresearch impacting not only the way we write code with conversational agents\nlike OpenAI's ChatGPT, Google's Bard, or Anthropic's Claude, the way we\ntranslate code from one language into another, but also the compiler\ninfrastructure underlying the language. While modeling approaches may vary and\nrepresentations differ, the targeted tasks often remain the same within the\nindividual classes of models. Relying solely on the ability of modern models to\nextract information from unstructured code does not take advantage of 70 years\nof programming language and compiler development by not utilizing the structure\ninherent to programs in the data collection. This detracts from the performance\nof models working over a tokenized representation of input code and precludes\nthe use of these models in the compiler itself. To work towards the first\nintermediate representation (IR) based models, we fully utilize the LLVM\ncompiler infrastructure, shared by a number of languages, to generate a 182B\ntoken dataset of LLVM IR. We generated this dataset from programming languages\nbuilt on the shared LLVM infrastructure, including Rust, Swift, Julia, and\nC/C++, by hooking into LLVM code generation either through the language's\npackage manager or the compiler directly to extract the dataset of intermediate\nrepresentations from production grade programs. Statistical analysis proves the\nutility of our dataset not only for large language model training, but also for\nthe introspection into the code generation process itself with the dataset\nshowing great promise for machine-learned compiler components.", "title": "ComPile: A Large IR Dataset from Production Sources", "url": "http://arxiv.org/abs/2309.15432v1" }

{ "text_length": 1763 }

{ "abstract": "Distributional shift, or the mismatch between training and deployment data,\nis a significant obstacle to the usage of machine learning in high-stakes\nindustrial applications, such as autonomous driving and medicine. This creates\na need to be able to assess how robustly ML models generalize as well as the\nquality of their uncertainty estimates. Standard ML baseline datasets do not\nallow these properties to be assessed, as the training, validation and test\ndata are often identically distributed. Recently, a range of dedicated\nbenchmarks have appeared, featuring both distributionally matched and shifted\ndata. Among these benchmarks, the Shifts dataset stands out in terms of the\ndiversity of tasks as well as the data modalities it features. While most of\nthe benchmarks are heavily dominated by 2D image classification tasks, Shifts\ncontains tabular weather forecasting, machine translation, and vehicle motion\nprediction tasks. This enables the robustness properties of models to be\nassessed on a diverse set of industrial-scale tasks and either universal or\ndirectly applicable task-specific conclusions to be reached. In this paper, we\nextend the Shifts Dataset with two datasets sourced from industrial, high-risk\napplications of high societal importance. Specifically, we consider the tasks\nof segmentation of white matter Multiple Sclerosis lesions in 3D magnetic\nresonance brain images and the estimation of power consumption in marine cargo\nvessels. Both tasks feature ubiquitous distributional shifts and a strict\nsafety requirement due to the high cost of errors. These new datasets will\nallow researchers to further explore robust generalization and uncertainty\nestimation in new situations. In this work, we provide a description of the\ndataset and baseline results for both tasks.", "title": "Shifts 2.0: Extending The Dataset of Real Distributional Shifts", "url": "http://arxiv.org/abs/2206.15407v2" }

{ "text_length": 1897 }

{ "abstract": "We introduce the first large-scale dataset, MNISQ, for both the Quantum and\nthe Classical Machine Learning community during the Noisy Intermediate-Scale\nQuantum era. MNISQ consists of 4,950,000 data points organized in 9\nsubdatasets. Building our dataset from the quantum encoding of classical\ninformation (e.g., MNIST dataset), we deliver a dataset in a dual form: in\nquantum form, as circuits, and in classical form, as quantum circuit\ndescriptions (quantum programming language, QASM). In fact, also the Machine\nLearning research related to quantum computers undertakes a dual challenge:\nenhancing machine learning exploiting the power of quantum computers, while\nalso leveraging state-of-the-art classical machine learning methodologies to\nhelp the advancement of quantum computing. Therefore, we perform circuit\nclassification on our dataset, tackling the task with both quantum and\nclassical models. In the quantum endeavor, we test our circuit dataset with\nQuantum Kernel methods, and we show excellent results up to $97\\%$ accuracy. In\nthe classical world, the underlying quantum mechanical structures within the\nquantum circuit data are not trivial. Nevertheless, we test our dataset on\nthree classical models: Structured State Space sequence model (S4), Transformer\nand LSTM. In particular, the S4 model applied on the tokenized QASM sequences\nreaches an impressive $77\\%$ accuracy. These findings illustrate that quantum\ncircuit-related datasets are likely to be quantum advantageous, but also that\nstate-of-the-art machine learning methodologies can competently classify and\nrecognize quantum circuits. We finally entrust the quantum and classical\nmachine learning community the fundamental challenge to build more\nquantum-classical datasets like ours and to build future benchmarks from our\nexperiments. The dataset is accessible on GitHub and its circuits are easily\nrun in qulacs or qiskit.", "title": "MNISQ: A Large-Scale Quantum Circuit Dataset for Machine Learning on/for Quantum Computers in the NISQ era", "url": "http://arxiv.org/abs/2306.16627v1" }

{ "text_length": 2046 }

{ "abstract": "Training of Machine Learning (ML) models in real contexts often deals with\nbig data sets and high-class imbalance samples where the class of interest is\nunrepresented (minority class). Practical solutions using classical ML models\naddress the problem of large data sets using parallel/distributed\nimplementations of training algorithms, approximate model-based solutions, or\napplying instance selection (IS) algorithms to eliminate redundant information.\nHowever, the combined problem of big and high imbalanced datasets has been less\naddressed. This work proposes three new methods for IS to be able to deal with\nlarge and imbalanced data sets. The proposed methods use Locality Sensitive\nHashing (LSH) as a base clustering technique, and then three different sampling\nmethods are applied on top of the clusters (or buckets) generated by LSH. The\nalgorithms were developed in the Apache Spark framework, guaranteeing their\nscalability. The experiments carried out in three different datasets suggest\nthat the proposed IS methods can improve the performance of a base ML model\nbetween 5% and 19% in terms of the geometric mean.", "title": "An Instance Selection Algorithm for Big Data in High imbalanced datasets based on LSH", "url": "http://arxiv.org/abs/2210.04310v1" }

{ "text_length": 1247 }

{ "abstract": "Agriculture is of one of the few remaining sectors that is yet to receive\nproper attention from the machine learning community. The importance of\ndatasets in the machine learning discipline cannot be overemphasized. The lack\nof standard and publicly available datasets related to agriculture impedes\npractitioners of this discipline to harness the full benefit of these powerful\ncomputational predictive tools and techniques. To improve this scenario, we\ndevelop, to the best of our knowledge, the first-ever standard, ready-to-use,\nand publicly available dataset of mango leaves. The images are collected from\nfour mango orchards of Bangladesh, one of the top mango-growing countries of\nthe world. The dataset contains 4000 images of about 1800 distinct leaves\ncovering seven diseases. Although the dataset is developed using mango leaves\nof Bangladesh only, since we deal with diseases that are common across many\ncountries, this dataset is likely to be applicable to identify mango diseases\nin other countries as well, thereby boosting mango yield. This dataset is\nexpected to draw wide attention from machine learning researchers and\npractitioners in the field of automated agriculture.", "title": "MangoLeafBD: A Comprehensive Image Dataset to Classify Diseased and Healthy Mango Leaves", "url": "http://arxiv.org/abs/2209.02377v1" }

{ "text_length": 1313 }

{ "abstract": "From face recognition in smartphones to automatic routing on self-driving\ncars, machine vision algorithms lie in the core of these features. These\nsystems solve image based tasks by identifying and understanding objects,\nsubsequently making decisions from these information. However, errors in\ndatasets are usually induced or even magnified in algorithms, at times\nresulting in issues such as recognising black people as gorillas and\nmisrepresenting ethnicities in search results. This paper tracks the errors in\ndatasets and their impacts, revealing that a flawed dataset could be a result\nof limited categories, incomprehensive sourcing and poor classification.", "title": "Seeing the Unseen: Errors and Bias in Visual Datasets", "url": "http://arxiv.org/abs/2211.01847v1" }

{ "text_length": 751 }

{ "abstract": "Despite the large amount of brain CT data generated in clinical practice, the\navailability of CT datasets for deep learning (DL) research is currently\nlimited. Furthermore, the data can be insufficiently or improperly prepared for\nmachine learning and thus lead to spurious and irreproducible analyses. This\nlack of access to comprehensive and diverse datasets poses a significant\nchallenge for the development of DL algorithms. In this work, we propose a\ncomplete semi-automatic pipeline to address the challenges of preparing a\nclinical brain CT dataset for DL analysis and describe the process of\nstandardising this heterogeneous dataset. Challenges include handling image\nsets with different orientations (axial, sagittal, coronal), different image\ntypes (to view soft tissues or bones) and dimensions, and removing redundant\nbackground. The final pipeline was able to process 5,868/10,659 (45%) CT image\ndatasets. Reasons for rejection include non-axial data (n=1,920), bone\nreformats (n=687), separated skull base/vault images (n=1,226), and\nregistration failures (n=465). Further format adjustments, including image\ncropping, resizing and scaling are also needed for DL processing. Of the axial\nscans that were not localisers, bone reformats or split brains, 5,868/6,333\n(93%) were accepted, while the remaining 465 failed the registration process.\nAppropriate preparation of medical imaging datasets for DL is a costly and\ntime-intensive process.", "title": "Challenges of building medical image datasets for development of deep learning software in stroke", "url": "http://arxiv.org/abs/2309.15081v1" }

{ "text_length": 1586 }

{ "abstract": "Machine learning approaches often require training and evaluation datasets\nwith a clear separation between positive and negative examples. This risks\nsimplifying and even obscuring the inherent subjectivity present in many tasks.\nPreserving such variance in content and diversity in datasets is often\nexpensive and laborious. This is especially troubling when building safety\ndatasets for conversational AI systems, as safety is both socially and\nculturally situated. To demonstrate this crucial aspect of conversational AI\nsafety, and to facilitate in-depth model performance analyses, we introduce the\nDICES (Diversity In Conversational AI Evaluation for Safety) dataset that\ncontains fine-grained demographic information about raters, high replication of\nratings per item to ensure statistical power for analyses, and encodes rater\nvotes as distributions across different demographics to allow for in-depth\nexplorations of different aggregation strategies. In short, the DICES dataset\nenables the observation and measurement of variance, ambiguity, and diversity\nin the context of conversational AI safety. We also illustrate how the dataset\noffers a basis for establishing metrics to show how raters' ratings can\nintersects with demographic categories such as racial/ethnic groups, age\ngroups, and genders. The goal of DICES is to be used as a shared resource and\nbenchmark that respects diverse perspectives during safety evaluation of\nconversational AI systems.", "title": "DICES Dataset: Diversity in Conversational AI Evaluation for Safety", "url": "http://arxiv.org/abs/2306.11247v1" }

{ "text_length": 1569 }

{ "abstract": "The recent and increasing interest in video-language research has driven the\ndevelopment of large-scale datasets that enable data-intensive machine learning\ntechniques. In comparison, limited effort has been made at assessing the\nfitness of these datasets for the video-language grounding task. Recent works\nhave begun to discover significant limitations in these datasets, suggesting\nthat state-of-the-art techniques commonly overfit to hidden dataset biases. In\nthis work, we present MAD (Movie Audio Descriptions), a novel benchmark that\ndeparts from the paradigm of augmenting existing video datasets with text\nannotations and focuses on crawling and aligning available audio descriptions\nof mainstream movies. MAD contains over 384,000 natural language sentences\ngrounded in over 1,200 hours of videos and exhibits a significant reduction in\nthe currently diagnosed biases for video-language grounding datasets. MAD's\ncollection strategy enables a novel and more challenging version of\nvideo-language grounding, where short temporal moments (typically seconds long)\nmust be accurately grounded in diverse long-form videos that can last up to\nthree hours. We have released MAD's data and baselines code at\nhttps://github.com/Soldelli/MAD.", "title": "MAD: A Scalable Dataset for Language Grounding in Videos from Movie Audio Descriptions", "url": "http://arxiv.org/abs/2112.00431v2" }

{ "text_length": 1363 }

{ "abstract": "In the current network-based computing world, where the number of\ninterconnected devices grows exponentially, their diversity, malfunctions, and\ncybersecurity threats are increasing at the same rate. To guarantee the correct\nfunctioning and performance of novel environments such as Smart Cities,\nIndustry 4.0, or crowdsensing, it is crucial to identify the capabilities of\ntheir devices (e.g., sensors, actuators) and detect potential misbehavior that\nmay arise due to cyberattacks, system faults, or misconfigurations. With this\ngoal in mind, a promising research field emerged focusing on creating and\nmanaging fingerprints that model the behavior of both the device actions and\nits components. The article at hand studies the recent growth of the device\nbehavior fingerprinting field in terms of application scenarios, behavioral\nsources, and processing and evaluation techniques. First, it performs a\ncomprehensive review of the device types, behavioral data, and processing and\nevaluation techniques used by the most recent and representative research works\ndealing with two major scenarios: device identification and device misbehavior\ndetection. After that, each work is deeply analyzed and compared, emphasizing\nits characteristics, advantages, and limitations. This article also provides\nresearchers with a review of the most relevant characteristics of existing\ndatasets as most of the novel processing techniques are based on machine\nlearning and deep learning. Finally, it studies the evolution of these two\nscenarios in recent years, providing lessons learned, current trends, and\nfuture research challenges to guide new solutions in the area.", "title": "A Survey on Device Behavior Fingerprinting: Data Sources, Techniques, Application Scenarios, and Datasets", "url": "http://arxiv.org/abs/2008.03343v2" }

{ "text_length": 1797 }

{ "abstract": "We introduce a pattern mining framework that operates on semi-structured\ndatasets and exploits the dichotomy between outcomes. Our approach takes\nadvantage of constraint reasoning to find sequential patterns that occur\nfrequently and exhibit desired properties. This allows the creation of novel\npattern embeddings that are useful for knowledge extraction and predictive\nmodeling. Finally, we present an application on customer intent prediction from\ndigital clickstream data. Overall, we show that pattern embeddings play an\nintegrator role between semi-structured data and machine learning models,\nimprove the performance of the downstream task and retain interpretability.", "title": "Dichotomic Pattern Mining with Applications to Intent Prediction from Semi-Structured Clickstream Datasets", "url": "http://arxiv.org/abs/2201.09178v1" }

{ "text_length": 816 }

{ "abstract": "The development of semi-supervised learning techniques is essential to\nenhance the generalization capacities of machine learning algorithms. Indeed,\nraw image data are abundant while labels are scarce, therefore it is crucial to\nleverage unlabeled inputs to build better models. The availability of large\ndatabases have been key for the development of learning algorithms with high\nlevel performance.\n Despite the major role of machine learning in Earth Observation to derive\nproducts such as land cover maps, datasets in the field are still limited,\neither because of modest surface coverage, lack of variety of scenes or\nrestricted classes to identify. We introduce a novel large-scale dataset for\nsemi-supervised semantic segmentation in Earth Observation, the MiniFrance\nsuite. MiniFrance has several unprecedented properties: it is large-scale,\ncontaining over 2000 very high resolution aerial images, accounting for more\nthan 200 billions samples (pixels); it is varied, covering 16 conurbations in\nFrance, with various climates, different landscapes, and urban as well as\ncountryside scenes; and it is challenging, considering land use classes with\nhigh-level semantics. Nevertheless, the most distinctive quality of MiniFrance\nis being the only dataset in the field especially designed for semi-supervised\nlearning: it contains labeled and unlabeled images in its training partition,\nwhich reproduces a life-like scenario. Along with this dataset, we present\ntools for data representativeness analysis in terms of appearance similarity\nand a thorough study of MiniFrance data, demonstrating that it is suitable for\nlearning and generalizes well in a semi-supervised setting. Finally, we present\nsemi-supervised deep architectures based on multi-task learning and the first\nexperiments on MiniFrance.", "title": "Semi-Supervised Semantic Segmentation in Earth Observation: The MiniFrance Suite, Dataset Analysis and Multi-task Network Study", "url": "http://arxiv.org/abs/2010.07830v1" }

{ "text_length": 1970 }

{ "abstract": "State recognition in well-known and customizable environments such as\nvehicles enables novel insights into users and potentially their intentions.\nBesides safety-relevant insights into, for example, fatigue, user\nexperience-related assessments become increasingly relevant. As thermal comfort\nis vital for overall comfort, we introduce a dataset for its prediction in\nvehicles incorporating 31 input signals and self-labeled user ratings based on\na 7-point Likert scale (-3 to +3) by 21 subjects. An importance ranking of such\nsignals indicates higher impact on prediction for signals like ambient\ntemperature, ambient humidity, radiation temperature, and skin temperature.\nLeveraging modern machine learning architectures enables us to not only\nautomatically recognize human thermal comfort state but also predict future\nstates. We provide details on how we train a recurrent network-based classifier\nand, thus, perform an initial performance benchmark of our proposed thermal\ncomfort dataset. Ultimately, we compare our collected dataset to publicly\navailable datasets.", "title": "AutoTherm: A Dataset and Ablation Study for Thermal Comfort Prediction in Vehicles", "url": "http://arxiv.org/abs/2211.08257v2" }

{ "text_length": 1188 }

{ "abstract": "Recent research in computational imaging largely focuses on developing\nmachine learning (ML) techniques for image reconstruction, which requires\nlarge-scale training datasets consisting of measurement data and ground-truth\nimages. However, suitable experimental datasets for X-ray Computed Tomography\n(CT) are scarce, and methods are often developed and evaluated only on\nsimulated data. We fill this gap by providing the community with a versatile,\nopen 2D fan-beam CT dataset suitable for developing ML techniques for a range\nof image reconstruction tasks. To acquire it, we designed a sophisticated,\nsemi-automatic scan procedure that utilizes a highly-flexible laboratory X-ray\nCT setup. A diverse mix of samples with high natural variability in shape and\ndensity was scanned slice-by-slice (5000 slices in total) with high angular and\nspatial resolution and three different beam characteristics: A high-fidelity, a\nlow-dose and a beam-hardening-inflicted mode. In addition, 750\nout-of-distribution slices were scanned with sample and beam variations to\naccommodate robustness and segmentation tasks. We provide raw projection data,\nreference reconstructions and segmentations based on an open-source data\nprocessing pipeline.", "title": "2DeteCT -- A large 2D expandable, trainable, experimental Computed Tomography dataset for machine learning", "url": "http://arxiv.org/abs/2306.05907v1" }

{ "text_length": 1371 }

{ "abstract": "In this paper, we introduce MIMII DUE, a new dataset for malfunctioning\nindustrial machine investigation and inspection with domain shifts due to\nchanges in operational and environmental conditions. Conventional methods for\nanomalous sound detection face practical challenges because the distribution of\nfeatures changes between the training and operational phases (called domain\nshift) due to various real-world factors. To check the robustness against\ndomain shifts, we need a dataset that actually includes domain shifts, but such\na dataset does not exist so far. The new dataset we created consists of the\nnormal and abnormal operating sounds of five different types of industrial\nmachines under two different operational/environmental conditions (source\ndomain and target domain) independent of normal/abnormal, with domain shifts\noccurring between the two domains. Experimental results showed significant\nperformance differences between the source and target domains, indicating that\nthe dataset contains the domain shifts. These findings demonstrate that the\ndataset will be helpful for checking the robustness against domain shifts. The\ndataset is a subset of the dataset for DCASE 2021 Challenge Task 2 and freely\navailable for download at https://zenodo.org/record/4740355", "title": "MIMII DUE: Sound Dataset for Malfunctioning Industrial Machine Investigation and Inspection with Domain Shifts due to Changes in Operational and Environmental Conditions", "url": "http://arxiv.org/abs/2105.02702v3" }

{ "text_length": 1486 }

{ "abstract": "Human annotations play a crucial role in machine learning (ML) research and\ndevelopment. However, the ethical considerations around the processes and\ndecisions that go into building ML datasets has not received nearly enough\nattention. In this paper, we survey an array of literature that provides\ninsights into ethical considerations around crowdsourced dataset annotation. We\nsynthesize these insights, and lay out the challenges in this space along two\nlayers: (1) who the annotator is, and how the annotators' lived experiences can\nimpact their annotations, and (2) the relationship between the annotators and\nthe crowdsourcing platforms and what that relationship affords them. Finally,\nwe put forth a concrete set of recommendations and considerations for dataset\ndevelopers at various stages of the ML data pipeline: task formulation,\nselection of annotators, platform and infrastructure choices, dataset analysis\nand evaluation, and dataset documentation and release.", "title": "Whose Ground Truth? Accounting for Individual and Collective Identities Underlying Dataset Annotation", "url": "http://arxiv.org/abs/2112.04554v1" }

{ "text_length": 1111 }

{ "abstract": "With the growing rates of cyber-attacks and cyber espionage, the need for\nbetter and more powerful intrusion detection systems (IDS) is even more\nwarranted nowadays. The basic task of an IDS is to act as the first line of\ndefense, in detecting attacks on the internet. As intrusion tactics from\nintruders become more sophisticated and difficult to detect, researchers have\nstarted to apply novel Machine Learning (ML) techniques to effectively detect\nintruders and hence preserve internet users' information and overall trust in\nthe entire internet network security. Over the last decade, there has been an\nexplosion of research on intrusion detection techniques based on ML and Deep\nLearning (DL) architectures on various cyber security-based datasets such as\nthe DARPA, KDDCUP'99, NSL-KDD, CAIDA, CTU-13, UNSW-NB15. In this research, we\nreview contemporary literature and provide a comprehensive survey of different\ntypes of intrusion detection technique that applies Support Vector Machines\n(SVMs) algorithms as a classifier. We focus only on studies that have been\nevaluated on the two most widely used datasets in cybersecurity namely: the\nKDDCUP'99 and the NSL-KDD datasets. We provide a summary of each method,\nidentifying the role of the SVMs classifier, and all other algorithms involved\nin the studies. Furthermore, we present a critical review of each method, in\ntabular form, highlighting the performance measures, strengths, and limitations\nof each of the methods surveyed.", "title": "Intrusion Detection Systems Using Support Vector Machines on the KDDCUP'99 and NSL-KDD Datasets: A Comprehensive Survey", "url": "http://arxiv.org/abs/2209.05579v1" }

{ "text_length": 1640 }

{ "abstract": "Cerebral stroke, the second most substantial cause of death universally, has\nbeen a primary public health concern over the last few years. With the help of\nmachine learning techniques, early detection of various stroke alerts is\naccessible, which can efficiently prevent or diminish the stroke. Medical\ndataset, however, are frequently unbalanced in their class label, with a\ntendency to poorly predict minority classes. In this paper, the potential risk\nfactors for stroke are investigated. Moreover, four distinctive approaches are\napplied to improve the classification of the minority class in the imbalanced\nstroke dataset, which are the ensemble weight voting classifier, the Synthetic\nMinority Over-sampling Technique (SMOTE), Principal Component Analysis with\nK-Means Clustering (PCA-Kmeans), Focal Loss with the Deep Neural Network (DNN)\nand compare their performance. Through the analysis results, SMOTE and\nPCA-Kmeans with DNN-Focal Loss work best for the limited size of a large severe\nimbalanced dataset,which is 2-4 times outperform Kaggle work.", "title": "Machine Learning Performance Analysis to Predict Stroke Based on Imbalanced Medical Dataset", "url": "http://arxiv.org/abs/2211.07652v1" }

{ "text_length": 1184 }

{ "abstract": "Delirium occurs in about 80% cases in the Intensive Care Unit (ICU) and is\nassociated with a longer hospital stay, increased mortality and other related\nissues. Delirium does not have any biomarker-based diagnosis and is commonly\ntreated with antipsychotic drugs (APD). However, multiple studies have shown\ncontroversy over the efficacy or safety of APD in treating delirium. Since\nrandomized controlled trials (RCT) are costly and time-expensive, we aim to\napproach the research question of the efficacy of APD in the treatment of\ndelirium using retrospective cohort analysis. We plan to use the Causal\ninference framework to look for the underlying causal structure model,\nleveraging the availability of large observational data on ICU patients. To\nexplore safety outcomes associated with APD, we aim to build a causal model for\ndelirium in the ICU using large observational data sets connecting various\ncovariates correlated with delirium. We utilized the MIMIC III database, an\nextensive electronic health records (EHR) dataset with 53,423 distinct hospital\nadmissions. Our null hypothesis is: there is no significant difference in\noutcomes for delirium patients under different drug-group in the ICU. Through\nour exploratory, machine learning based and causal analysis, we had findings\nsuch as: mean length-of-stay and max length-of-stay is higher for patients in\nHaloperidol drug group, and haloperidol group has a higher rate of death in a\nyear compared to other two-groups. Our generated causal model explicitly shows\nthe functional relationships between different covariates. For future work, we\nplan to do time-varying analysis on the dataset.", "title": "Causal Discovery on the Effect of Antipsychotic Drugs on Delirium Patients in the ICU using Large EHR Dataset", "url": "http://arxiv.org/abs/2205.01057v1" }

{ "text_length": 1797 }

{ "abstract": "In order to simulate human language capacity, natural language processing\nsystems must be able to reason about the dynamics of everyday situations,\nincluding their possible causes and effects. Moreover, they should be able to\ngeneralise the acquired world knowledge to new languages, modulo cultural\ndifferences. Advances in machine reasoning and cross-lingual transfer depend on\nthe availability of challenging evaluation benchmarks. Motivated by both\ndemands, we introduce Cross-lingual Choice of Plausible Alternatives (XCOPA), a\ntypologically diverse multilingual dataset for causal commonsense reasoning in\n11 languages, which includes resource-poor languages like Eastern Apur\\'imac\nQuechua and Haitian Creole. We evaluate a range of state-of-the-art models on\nthis novel dataset, revealing that the performance of current methods based on\nmultilingual pretraining and zero-shot fine-tuning falls short compared to\ntranslation-based transfer. Finally, we propose strategies to adapt\nmultilingual models to out-of-sample resource-lean languages where only a small\ncorpus or a bilingual dictionary is available, and report substantial\nimprovements over the random baseline. The XCOPA dataset is freely available at\ngithub.com/cambridgeltl/xcopa.", "title": "XCOPA: A Multilingual Dataset for Causal Commonsense Reasoning", "url": "http://arxiv.org/abs/2005.00333v2" }

{ "text_length": 1346 }

{ "abstract": "In a depression-diagnosis-directed clinical session, doctors initiate a\nconversation with ample emotional support that guides the patients to expose\ntheir symptoms based on clinical diagnosis criteria. Such a dialogue system is\ndistinguished from existing single-purpose human-machine dialog systems, as it\ncombines task-oriented and chit-chats with uniqueness in dialogue topics and\nprocedures. However, due to the social stigma associated with mental illness,\nthe dialogue data related to depression consultation and diagnosis are rarely\ndisclosed. Based on clinical depression diagnostic criteria ICD-11 and DSM-5,\nwe designed a 3-phase procedure to construct D$^4$: a Chinese Dialogue Dataset\nfor Depression-Diagnosis-Oriented Chat, which simulates the dialogue between\ndoctors and patients during the diagnosis of depression, including diagnosis\nresults and symptom summary given by professional psychiatrists for each\nconversation. Upon the newly-constructed dataset, four tasks mirroring the\ndepression diagnosis process are established: response generation, topic\nprediction, dialog summary, and severity classification of depressive episode\nand suicide risk. Multi-scale evaluation results demonstrate that a more\nempathy-driven and diagnostic-accurate consultation dialogue system trained on\nour dataset can be achieved compared to rule-based bots.", "title": "D4: a Chinese Dialogue Dataset for Depression-Diagnosis-Oriented Chat", "url": "http://arxiv.org/abs/2205.11764v2" }

{ "text_length": 1462 }

{ "abstract": "Quantum and quantum-inspired machine learning has emerged as a promising and\nchallenging research field due to the increased popularity of quantum\ncomputing, especially with near-term devices. Theoretical contributions point\ntoward generative modeling as a promising direction to realize the first\nexamples of real-world quantum advantages from these technologies. A few\nempirical studies also demonstrate such potential, especially when considering\nquantum-inspired models based on tensor networks. In this work, we apply\ntensor-network-based generative models to the problem of molecular discovery.\nIn our approach, we utilize two small molecular datasets: a subset of $4989$\nmolecules from the QM9 dataset and a small in-house dataset of $516$ validated\nantioxidants from TotalEnergies. We compare several tensor network models\nagainst a generative adversarial network using different sample-based metrics,\nwhich reflect their learning performances on each task, and multiobjective\nperformances using $3$ relevant molecular metrics per task. We also combined\nthe output of the models and demonstrate empirically that such a combination\ncan be beneficial, advocating for the unification of classical and\nquantum(-inspired) generative learning.", "title": "Application of quantum-inspired generative models to small molecular datasets", "url": "http://arxiv.org/abs/2304.10867v1" }

{ "text_length": 1357 }

{ "abstract": "Data imbalance is a well-known issue in the field of machine learning,\nattributable to the cost of data collection, the difficulty of labeling, and\nthe geographical distribution of the data. In computer vision, bias in data\ndistribution caused by image appearance remains highly unexplored. Compared to\ncategorical distributions using class labels, image appearance reveals complex\nrelationships between objects beyond what class labels provide. Clustering deep\nperceptual features extracted from raw pixels gives a richer representation of\nthe data. This paper presents a novel method for addressing data imbalance in\nmachine learning. The method computes sample likelihoods based on image\nappearance using deep perceptual embeddings and clustering. It then uses these\nlikelihoods to weigh samples differently during training with a proposed\n$\\textbf{Generalized Focal Loss}$ function. This loss can be easily integrated\nwith deep learning algorithms. Experiments validate the method's effectiveness\nacross autonomous driving vision datasets including KITTI and nuScenes. The\nloss function improves state-of-the-art 3D object detection methods, achieving\nover $200\\%$ AP gains on under-represented classes (Cyclist) in the KITTI\ndataset. The results demonstrate the method is generalizable, complements\nexisting techniques, and is particularly beneficial for smaller datasets and\nrare classes. Code is available at:\nhttps://github.com/towardsautonomy/DatasetEquity", "title": "DatasetEquity: Are All Samples Created Equal? In The Quest For Equity Within Datasets", "url": "http://arxiv.org/abs/2308.09878v2" }

{ "text_length": 1585 }

{ "abstract": "The availability of different pre-trained semantic models enabled the quick\ndevelopment of machine learning components for downstream applications. Despite\nthe availability of abundant text data for low resource languages, only a few\nsemantic models are publicly available. Publicly available pre-trained models\nare usually built as a multilingual version of semantic models that can not fit\nwell for each language due to context variations. In this work, we introduce\ndifferent semantic models for Amharic. After we experiment with the existing\npre-trained semantic models, we trained and fine-tuned nine new different\nmodels using a monolingual text corpus. The models are build using word2Vec\nembeddings, distributional thesaurus (DT), contextual embeddings, and DT\nembeddings obtained via network embedding algorithms. Moreover, we employ these\nmodels for different NLP tasks and investigate their impact. We find that newly\ntrained models perform better than pre-trained multilingual models.\nFurthermore, models based on contextual embeddings from RoBERTA perform better\nthan the word2Vec models.", "title": "Introducing various Semantic Models for Amharic: Experimentation and Evaluation with multiple Tasks and Datasets", "url": "http://arxiv.org/abs/2011.01154v2" }

{ "text_length": 1248 }

{ "abstract": "WeatherBench is a benchmark dataset for medium-range weather forecasting of\ngeopotential, temperature and precipitation, consisting of preprocessed data,\npredefined evaluation metrics and a number of baseline models. WeatherBench\nProbability extends this to probabilistic forecasting by adding a set of\nestablished probabilistic verification metrics (continuous ranked probability\nscore, spread-skill ratio and rank histograms) and a state-of-the-art\noperational baseline using the ECWMF IFS ensemble forecast. In addition, we\ntest three different probabilistic machine learning methods -- Monte Carlo\ndropout, parametric prediction and categorical prediction, in which the\nprobability distribution is discretized. We find that plain Monte Carlo dropout\nseverely underestimates uncertainty. The parametric and categorical models both\nproduce fairly reliable forecasts of similar quality. The parametric models\nhave fewer degrees of freedom while the categorical model is more flexible when\nit comes to predicting non-Gaussian distributions. None of the models are able\nto match the skill of the operational IFS model. We hope that this benchmark\nwill enable other researchers to evaluate their probabilistic approaches.", "title": "WeatherBench Probability: A benchmark dataset for probabilistic medium-range weather forecasting along with deep learning baseline models", "url": "http://arxiv.org/abs/2205.00865v1" }

{ "text_length": 1391 }

{ "abstract": "We present a physically motivated strategy for the construction of training\nsets for transferable machine learning interatomic potentials. It is based on a\nsystematic exploration of all possible space groups in random crystal\nstructures, together with deformations of cell shape, size, and atomic\npositions. The resulting potentials turn out to be unbiased and generically\napplicable to studies of bulk defects without including any defect structures\nin the training set or employing any additional Active Learning. Using this\napproach we construct transferable potentials for pure Magnesium that reproduce\nthe properties of hexagonal closed packed (hcp) and body centered cubic (bcc)\npolymorphs very well. In the process we investigate how different types of\ntraining structures impact the properties and the predictive power of the\nresulting potential.", "title": "Systematic Atomic Structure Datasets for Machine Learning Potentials: Application to Defects in Magnesium", "url": "http://arxiv.org/abs/2207.04009v4" }

{ "text_length": 994 }

{ "abstract": "Materials datasets are usually featured by the existence of many redundant\n(highly similar) materials due to the tinkering material design practice over\nthe history of materials research. For example, the materials project database\nhas many perovskite cubic structure materials similar to SrTiO$_3$. This sample\nredundancy within the dataset makes the random splitting of machine learning\nmodel evaluation to fail so that the ML models tend to achieve over-estimated\npredictive performance which is misleading for the materials science community.\nThis issue is well known in the field of bioinformatics for protein function\nprediction, in which a redundancy reduction procedure (CD-Hit) is always\napplied to reduce the sample redundancy by ensuring no pair of samples has a\nsequence similarity greater than a given threshold. This paper surveys the\noverestimated ML performance in the literature for both composition based and\nstructure based material property prediction. We then propose a material\ndataset redundancy reduction algorithm called MD-HIT and evaluate it with\nseveral composition and structure based distance threshold sfor reducing data\nset sample redundancy. We show that with this control, the predicted\nperformance tends to better reflect their true prediction capability. Our\nMD-hit code can be freely accessed at https://github.com/usccolumbia/MD-HIT", "title": "MD-HIT: Machine learning for materials property prediction with dataset redundancy control", "url": "http://arxiv.org/abs/2307.04351v1" }

{ "text_length": 1495 }

{ "abstract": "The sensitivity of imaging spectroscopy to haemoglobin derivatives makes it a\npromising tool for detecting blood. However, due to complexity and high\ndimensionality of hyperspectral images, the development of hyperspectral blood\ndetection algorithms is challenging. To facilitate their development, we\npresent a new hyperspectral blood detection dataset. This dataset, published in\naccordance to open access mandate, consist of multiple detection scenarios with\nvarying levels of complexity. It allows to test the performance of Machine\nLearning methods in relation to different acquisition environments, types of\nbackground, age of blood and presence of other blood-like substances. We\nexplored the dataset with blood detection experiments. We used hyperspectral\ntarget detection algorithm based on the well-known Matched Filter detector. Our\nresults and their discussion highlight the challenges of blood detection in\nhyperspectral data and form a reference for further works.", "title": "A Dataset for Evaluating Blood Detection in Hyperspectral Images", "url": "http://arxiv.org/abs/2008.10254v2" }

{ "text_length": 1077 }

{ "abstract": "Magnetic particle imaging (MPI) is an imaging modality exploiting the\nnonlinear magnetization behavior of (super-)paramagnetic nanoparticles to\nobtain a space- and often also time-dependent concentration of a tracer\nconsisting of these nanoparticles. MPI has a continuously increasing number of\npotential medical applications. One prerequisite for successful performance in\nthese applications is a proper solution to the image reconstruction problem.\nMore classical methods from inverse problems theory, as well as novel\napproaches from the field of machine learning, have the potential to deliver\nhigh-quality reconstructions in MPI. We investigate a novel reconstruction\napproach based on a deep image prior, which builds on representing the solution\nby a deep neural network. Novel approaches, as well as variational and\niterative regularization techniques, are compared quantitatively in terms of\npeak signal-to-noise ratios and structural similarity indices on the publicly\navailable Open MPI dataset.", "title": "Deep image prior for 3D magnetic particle imaging: A quantitative comparison of regularization techniques on Open MPI dataset", "url": "http://arxiv.org/abs/2007.01593v1" }

{ "text_length": 1166 }

{ "abstract": "Misinformation during pandemic situations like COVID-19 is growing rapidly on\nsocial media and other platforms. This expeditious growth of misinformation\ncreates adverse effects on the people living in the society. Researchers are\ntrying their best to mitigate this problem using different approaches based on\nMachine Learning (ML), Deep Learning (DL), and Natural Language Processing\n(NLP). This survey aims to study different approaches of misinformation\ndetection on COVID-19 in recent literature to help the researchers in this\ndomain. More specifically, we review the different methods used for COVID-19\nmisinformation detection in their research with an overview of data\npre-processing and feature extraction methods to get a better understanding of\ntheir work. We also summarize the existing datasets which can be used for\nfurther research. Finally, we discuss the limitations of the existing methods\nand highlight some potential future research directions along this dimension to\ncombat the spreading of misinformation during a pandemic.", "title": "A Survey of COVID-19 Misinformation: Datasets, Detection Techniques and Open Issues", "url": "http://arxiv.org/abs/2110.00737v2" }

{ "text_length": 1163 }

{ "abstract": "The production, shipping, usage, and disposal of consumer goods have a\nsubstantial impact on greenhouse gas emissions and the depletion of resources.\nMachine Learning (ML) can help to foster sustainable consumption patterns by\naccounting for sustainability aspects in product search or recommendations of\nmodern retail platforms. However, the lack of large high quality publicly\navailable product data with trustworthy sustainability information impedes the\ndevelopment of ML technology that can help to reach our sustainability goals.\nHere we present GreenDB, a database that collects products from European online\nshops on a weekly basis. As proxy for the products' sustainability, it relies\non sustainability labels, which are evaluated by experts. The GreenDB schema\nextends the well-known schema.org Product definition and can be readily\nintegrated into existing product catalogs. We present initial results\ndemonstrating that ML models trained with our data can reliably (F1 score 96%)\npredict the sustainability label of products. These contributions can help to\ncomplement existing e-commerce experiences and ultimately encourage users to\nmore sustainable consumption patterns.", "title": "GreenDB -- A Dataset and Benchmark for Extraction of Sustainability Information of Consumer Goods", "url": "http://arxiv.org/abs/2207.10733v3" }

{ "text_length": 1317 }

{ "abstract": "We propose CodeQA, a free-form question answering dataset for the purpose of\nsource code comprehension: given a code snippet and a question, a textual\nanswer is required to be generated. CodeQA contains a Java dataset with 119,778\nquestion-answer pairs and a Python dataset with 70,085 question-answer pairs.\nTo obtain natural and faithful questions and answers, we implement syntactic\nrules and semantic analysis to transform code comments into question-answer\npairs. We present the construction process and conduct systematic analysis of\nour dataset. Experiment results achieved by several neural baselines on our\ndataset are shown and discussed. While research on question-answering and\nmachine reading comprehension develops rapidly, few prior work has drawn\nattention to code question answering. This new dataset can serve as a useful\nresearch benchmark for source code comprehension.", "title": "CodeQA: A Question Answering Dataset for Source Code Comprehension", "url": "http://arxiv.org/abs/2109.08365v1" }

{ "text_length": 990 }

{ "abstract": "Many statistical machine approaches could ultimately highlight novel features\nof the etiology of complex diseases by analyzing multi-omics data. However,\nthey are sensitive to some deviations in distribution when the observed samples\nare potentially contaminated with adversarial corrupted outliers (e.g., a\nfictional data distribution). Likewise, statistical advances lag in supporting\ncomprehensive data-driven analyses of complex multi-omics data integration. We\npropose a novel non-linear M-estimator-based approach, \"robust kernel machine\nregression (RobKMR),\" to improve the robustness of statistical machine\nregression and the diversity of fictional data to examine the higher-order\ncomposite effect of multi-omics datasets. We address a robust kernel-centered\nGram matrix to estimate the model parameters accurately. We also propose a\nrobust score test to assess the marginal and joint Hadamard product of features\nfrom multi-omics data. We apply our proposed approach to a multi-omics dataset\nof osteoporosis (OP) from Caucasian females. Experiments demonstrate that the\nproposed approach effectively identifies the inter-related risk factors of OP.\nWith solid evidence (p-value = 0.00001), biological validations, network-based\nanalysis, causal inference, and drug repurposing, the selected three triplets\n((DKK1, SMTN, DRGX), (MTND5, FASTKD2, CSMD3), (MTND5, COG3, CSMD3)) are\nsignificant biomarkers and directly relate to BMD. Overall, the top three\nselected genes (DKK1, MTND5, FASTKD2) and one gene (SIDT1 at p-value= 0.001)\nsignificantly bond with four drugs- Tacrolimus, Ibandronate, Alendronate, and\nBazedoxifene out of 30 candidates for drug repurposing in OP. Further, the\nproposed approach can be applied to any disease model where multi-omics\ndatasets are available.", "title": "A robust kernel machine regression towards biomarker selection in multi-omics datasets of osteoporosis for drug discovery", "url": "http://arxiv.org/abs/2201.05060v1" }

{ "text_length": 1943 }

{ "abstract": "Federated Learning (FL) is a novel approach enabling several clients holding\nsensitive data to collaboratively train machine learning models, without\ncentralizing data. The cross-silo FL setting corresponds to the case of few\n($2$--$50$) reliable clients, each holding medium to large datasets, and is\ntypically found in applications such as healthcare, finance, or industry. While\nprevious works have proposed representative datasets for cross-device FL, few\nrealistic healthcare cross-silo FL datasets exist, thereby slowing algorithmic\nresearch in this critical application. In this work, we propose a novel\ncross-silo dataset suite focused on healthcare, FLamby (Federated Learning\nAMple Benchmark of Your cross-silo strategies), to bridge the gap between\ntheory and practice of cross-silo FL. FLamby encompasses 7 healthcare datasets\nwith natural splits, covering multiple tasks, modalities, and data volumes,\neach accompanied with baseline training code. As an illustration, we\nadditionally benchmark standard FL algorithms on all datasets. Our flexible and\nmodular suite allows researchers to easily download datasets, reproduce results\nand re-use the different components for their research. FLamby is available\nat~\\url{www.github.com/owkin/flamby}.", "title": "FLamby: Datasets and Benchmarks for Cross-Silo Federated Learning in Realistic Healthcare Settings", "url": "http://arxiv.org/abs/2210.04620v3" }

{ "text_length": 1390 }

{ "abstract": "Twenty-five years ago, my colleagues Miyuki Kamachi and Jiro Gyoba and I\ndesigned and photographed JAFFE, a set of facial expression images intended for\nuse in a study of face perception. In 2019, without seeking permission or\ninforming us, Kate Crawford and Trevor Paglen exhibited JAFFE in two widely\npublicized art shows. In addition, they published a nonfactual account of the\nimages in the essay \"Excavating AI: The Politics of Images in Machine Learning\nTraining Sets.\" The present article recounts the creation of the JAFFE dataset\nand unravels each of Crawford and Paglen's fallacious statements. I also\ndiscuss JAFFE more broadly in connection with research on facial expression,\naffective computing, and human-computer interaction.", "title": "\"Excavating AI\" Re-excavated: Debunking a Fallacious Account of the JAFFE Dataset", "url": "http://arxiv.org/abs/2107.13998v1" }

{ "text_length": 857 }

{ "abstract": "Antrophonegic pressure (i.e. human influence) on the environment is one of\nthe largest causes of the loss of biological diversity. Wilderness areas, in\ncontrast, are home to undisturbed ecological processes. However, there is no\nbiophysical definition of the term wilderness. Instead, wilderness is more of a\nphilosophical or cultural concept and thus cannot be easily delineated or\ncategorized in a technical manner. With this paper, (i) we introduce the task\nof wilderness mapping by means of machine learning applied to satellite imagery\n(ii) and publish MapInWild, a large-scale benchmark dataset curated for that\ntask. MapInWild is a multi-modal dataset and comprises various geodata acquired\nand formed from a diverse set of Earth observation sensors. The dataset\nconsists of 8144 images with a shape of 1920 x 1920 pixels and is approximately\n350 GB in size. The images are weakly annotated with three classes derived from\nthe World Database of Protected Areas - Strict Nature Reserves, Wilderness\nAreas, and National Parks. With the dataset, which shall serve as a testbed for\ndevelopments in fields such as explainable machine learning and environmental\nremote sensing, we hope to contribute to a deepening of our understanding of\nthe question \"What makes nature wild?\".", "title": "MapInWild: A Remote Sensing Dataset to Address the Question What Makes Nature Wild", "url": "http://arxiv.org/abs/2212.02265v1" }

{ "text_length": 1396 }

{ "abstract": "Dataset Condensation is a newly emerging technique aiming at learning a tiny\ndataset that captures the rich information encoded in the original dataset. As\nthe size of datasets contemporary machine learning models rely on becomes\nincreasingly large, condensation methods become a prominent direction for\naccelerating network training and reducing data storage. Despite numerous\nmethods have been proposed in this rapidly growing field, evaluating and\ncomparing different condensation methods is non-trivial and still remains an\nopen issue. The quality of condensed dataset are often shadowed by many\ncritical contributing factors to the end performance, such as data augmentation\nand model architectures. The lack of a systematic way to evaluate and compare\ncondensation methods not only hinders our understanding of existing techniques,\nbut also discourages practical usage of the synthesized datasets. This work\nprovides the first large-scale standardized benchmark on Dataset Condensation.\nIt consists of a suite of evaluations to comprehensively reflect the\ngenerability and effectiveness of condensation methods through the lens of\ntheir generated dataset. Leveraging this benchmark, we conduct a large-scale\nstudy of current condensation methods, and report many insightful findings that\nopen up new possibilities for future development. The benchmark library,\nincluding evaluators, baseline methods, and generated datasets, is open-sourced\nto facilitate future research and application.", "title": "DC-BENCH: Dataset Condensation Benchmark", "url": "http://arxiv.org/abs/2207.09639v2" }

{ "text_length": 1568 }

{ "abstract": "Accurate bot detection is necessary for the safety and integrity of online\nplatforms. It is also crucial for research on the influence of bots in\nelections, the spread of misinformation, and financial market manipulation.\nPlatforms deploy infrastructure to flag or remove automated accounts, but their\ntools and data are not publicly available. Thus, the public must rely on\nthird-party bot detection. These tools employ machine learning and often\nachieve near perfect performance for classification on existing datasets,\nsuggesting bot detection is accurate, reliable and fit for use in downstream\napplications. We provide evidence that this is not the case and show that high\nperformance is attributable to limitations in dataset collection and labeling\nrather than sophistication of the tools. Specifically, we show that simple\ndecision rules -- shallow decision trees trained on a small number of features\n-- achieve near-state-of-the-art performance on most available datasets and\nthat bot detection datasets, even when combined together, do not generalize\nwell to out-of-sample datasets. Our findings reveal that predictions are highly\ndependent on each dataset's collection and labeling procedures rather than\nfundamental differences between bots and humans. These results have important\nimplications for both transparency in sampling and labeling procedures and\npotential biases in research using existing bot detection tools for\npre-processing.", "title": "Simplistic Collection and Labeling Practices Limit the Utility of Benchmark Datasets for Twitter Bot Detection", "url": "http://arxiv.org/abs/2301.07015v2" }

{ "text_length": 1598 }

{ "abstract": "Air quality forecasting has garnered significant attention recently, with\ndata-driven models taking center stage due to advancements in machine learning\nand deep learning models. However, researchers face challenges with complex\ndata acquisition and the lack of open-sourced datasets, hindering efficient\nmodel validation. This paper introduces PurpleAirSF, a comprehensive and easily\naccessible dataset collected from the PurpleAir network. With its high temporal\nresolution, various air quality measures, and diverse geographical coverage,\nthis dataset serves as a useful tool for researchers aiming to develop novel\nforecasting models, study air pollution patterns, and investigate their impacts\non health and the environment. We present a detailed account of the data\ncollection and processing methods employed to build PurpleAirSF. Furthermore,\nwe conduct preliminary experiments using both classic and modern\nspatio-temporal forecasting models, thereby establishing a benchmark for future\nair quality forecasting tasks.", "title": "Unleashing Realistic Air Quality Forecasting: Introducing the Ready-to-Use PurpleAirSF Dataset", "url": "http://arxiv.org/abs/2306.13948v1" }

{ "text_length": 1154 }

{ "abstract": "In this paper we present DISC, a dataset of millimeter-wave channel impulse\nresponse measurements for integrated human activity sensing and communication.\nThis is the first dataset collected with a software-defined radio testbed that\ntransmits 60 GHz IEEE 802-11ay-compliant packets and estimates the channel\nresponse including reflections of the signal on the moving body parts of\nsubjects moving in an indoor environment. The provided data contain the\ncontribution of 7 subjects performing 4 different activities. Differently from\navailable radar-based millimeter-wave sensing datasets, our measurements are\ncollected using both uniform packet transmission times and sparse traffic\npatterns from real Wi-Fi deployments. Thanks to these unique characteristics,\nDISC serves as a multi-purpose benchmarking tool for machine learning-based\nhuman activity recognition, radio frequency gait analysis, and sparse sensing\nalgorithms for next-generation integrated sensing and communication.", "title": "DISC: a Dataset for Integrated Sensing and Communication in mmWave Systems", "url": "http://arxiv.org/abs/2306.09469v1" }

{ "text_length": 1093 }

{ "abstract": "Malaria microscopy, microscopic examination of stained blood slides to detect\nparasite Plasmodium, is considered to be a gold-standard for detecting\nlife-threatening disease malaria. Detecting the plasmodium parasite requires a\nskilled examiner and may take up to 10 to 15 minutes to completely go through\nthe whole slide. Due to a lack of skilled medical professionals in the\nunderdeveloped or resource deficient regions, many cases go misdiagnosed;\nresulting in unavoidable complications and/or undue medication. We propose to\ncomplement the medical professionals by creating a deep learning-based method\nto automatically detect (localize) the plasmodium parasites in the photograph\nof stained film. To handle the unbalanced nature of the dataset, we adopt a\ntwo-stage approach. Where the first stage is trained to detect blood cells and\nclassify them into just healthy or infected. The second stage is trained to\nclassify each detected cell further into the life-cycle stage. To facilitate\nthe research in machine learning-based malaria microscopy, we introduce a new\nlarge scale microscopic image malaria dataset. Thirty-eight thousand cells are\ntagged from the 345 microscopic images of different Giemsa-stained slides of\nblood samples. Extensive experimentation is performed using different CNN\nbackbones including VGG, DenseNet, and ResNet on this dataset. Our experiments\nand analysis reveal that the two-stage approach works better than the one-stage\napproach for malaria detection. To ensure the usability of our approach, we\nhave also developed a mobile app that will be used by local hospitals for\ninvestigation and educational purposes. The dataset, its annotations, and\nimplementation codes will be released upon publication of the paper.", "title": "A Dataset and Benchmark for Malaria Life-Cycle Classification in Thin Blood Smear Images", "url": "http://arxiv.org/abs/2102.08708v1" }

{ "text_length": 1875 }

{ "abstract": "Precise hardware performance models play a crucial role in code\noptimizations. They can assist compilers in making heuristic decisions or aid\nautotuners in identifying the optimal configuration for a given program. For\nexample, the autotuner for XLA, a machine learning compiler, discovered 10-20%\nspeedup on state-of-the-art models serving substantial production traffic at\nGoogle. Although there exist a few datasets for program performance prediction,\nthey target small sub-programs such as basic blocks or kernels. This paper\nintroduces TpuGraphs, a performance prediction dataset on full tensor programs,\nrepresented as computational graphs, running on Tensor Processing Units (TPUs).\nEach graph in the dataset represents the main computation of a machine learning\nworkload, e.g., a training epoch or an inference step. Each data sample\ncontains a computational graph, a compilation configuration, and the execution\ntime of the graph when compiled with the configuration. The graphs in the\ndataset are collected from open-source machine learning programs, featuring\npopular model architectures, e.g., ResNet, EfficientNet, Mask R-CNN, and\nTransformer. TpuGraphs provides 25x more graphs than the largest graph property\nprediction dataset (with comparable graph sizes), and 770x larger graphs on\naverage compared to existing performance prediction datasets on machine\nlearning programs. This graph-level prediction task on large graphs introduces\nnew challenges in learning, ranging from scalability, training efficiency, to\nmodel quality.", "title": "TpuGraphs: A Performance Prediction Dataset on Large Tensor Computational Graphs", "url": "http://arxiv.org/abs/2308.13490v1" }

{ "text_length": 1658 }

{ "abstract": "The development of machine learning models for electrocatalysts requires a\nbroad set of training data to enable their use across a wide variety of\nmaterials. One class of materials that currently lacks sufficient training data\nis oxides, which are critical for the development of OER catalysts. To address\nthis, we developed the OC22 dataset, consisting of 62,331 DFT relaxations\n(~9,854,504 single point calculations) across a range of oxide materials,\ncoverages, and adsorbates. We define generalized total energy tasks that enable\nproperty prediction beyond adsorption energies; we test baseline performance of\nseveral graph neural networks; and we provide pre-defined dataset splits to\nestablish clear benchmarks for future efforts. In the most general task,\nGemNet-OC sees a ~36% improvement in energy predictions when combining the\nchemically dissimilar OC20 and OC22 datasets via fine-tuning. Similarly, we\nachieved a ~19% improvement in total energy predictions on OC20 and a ~9%\nimprovement in force predictions in OC22 when using joint training. We\ndemonstrate the practical utility of a top performing model by capturing\nliterature adsorption energies and important OER scaling relationships. We\nexpect OC22 to provide an important benchmark for models seeking to incorporate\nintricate long-range electrostatic and magnetic interactions in oxide surfaces.\nDataset and baseline models are open sourced, and a public leaderboard is\navailable to encourage continued community developments on the total energy\ntasks and data.", "title": "The Open Catalyst 2022 (OC22) Dataset and Challenges for Oxide Electrocatalysts", "url": "http://arxiv.org/abs/2206.08917v3" }

{ "text_length": 1646 }

{ "abstract": "Toxic language detection systems often falsely flag text that contains\nminority group mentions as toxic, as those groups are often the targets of\nonline hate. Such over-reliance on spurious correlations also causes systems to\nstruggle with detecting implicitly toxic language. To help mitigate these\nissues, we create ToxiGen, a new large-scale and machine-generated dataset of\n274k toxic and benign statements about 13 minority groups. We develop a\ndemonstration-based prompting framework and an adversarial\nclassifier-in-the-loop decoding method to generate subtly toxic and benign text\nwith a massive pretrained language model. Controlling machine generation in\nthis way allows ToxiGen to cover implicitly toxic text at a larger scale, and\nabout more demographic groups, than previous resources of human-written text.\nWe conduct a human evaluation on a challenging subset of ToxiGen and find that\nannotators struggle to distinguish machine-generated text from human-written\nlanguage. We also find that 94.5% of toxic examples are labeled as hate speech\nby human annotators. Using three publicly-available datasets, we show that\nfinetuning a toxicity classifier on our data improves its performance on\nhuman-written data substantially. We also demonstrate that ToxiGen can be used\nto fight machine-generated toxicity as finetuning improves the classifier\nsignificantly on our evaluation subset. Our code and data can be found at\nhttps://github.com/microsoft/ToxiGen.", "title": "ToxiGen: A Large-Scale Machine-Generated Dataset for Adversarial and Implicit Hate Speech Detection", "url": "http://arxiv.org/abs/2203.09509v4" }

{ "text_length": 1602 }

{ "abstract": "Human Activity Recognition (HAR) has recently received remarkable attention\nin numerous applications such as assisted living and remote monitoring.\nExisting solutions based on sensors and vision technologies have obtained\nachievements but still suffering from considerable limitations in the\nenvironmental requirement. Wireless signals like WiFi-based sensing have\nemerged as a new paradigm since it is convenient and not restricted in the\nenvironment. In this paper, a new WiFi-based and video-based neural network\n(WiNN) is proposed to improve the robustness of activity recognition where the\nsynchronized video serves as the supplement for the wireless data. Moreover, a\nwireless-vision benchmark (WiVi) is collected for 9 class actions recognition\nin three different visual conditions, including the scenes without occlusion,\nwith partial occlusion, and with full occlusion. Both machine learning methods\n- support vector machine (SVM) as well as deep learning methods are used for\nthe accuracy verification of the data set. Our results show that WiVi data set\nsatisfies the primary demand and all three branches in the proposed pipeline\nkeep more than $80\\%$ of activity recognition accuracy over multiple action\nsegmentation from 1s to 3s. In particular, WiNN is the most robust method in\nterms of all the actions on three action segmentation compared to the others.", "title": "A Wireless-Vision Dataset for Privacy Preserving Human Activity Recognition", "url": "http://arxiv.org/abs/2205.11962v1" }

{ "text_length": 1482 }

{ "abstract": "Falls are one of the leading cause of injury-related deaths among the elderly\nworldwide. Effective detection of falls can reduce the risk of complications\nand injuries. Fall detection can be performed using wearable devices or ambient\nsensors; these methods may struggle with user compliance issues or false\nalarms. Video cameras provide a passive alternative; however, regular RGB\ncameras are impacted by changing lighting conditions and privacy concerns. From\na machine learning perspective, developing an effective fall detection system\nis challenging because of the rarity and variability of falls. Many existing\nfall detection datasets lack important real-world considerations, such as\nvaried lighting, continuous activities of daily living (ADLs), and camera\nplacement. The lack of these considerations makes it difficult to develop\npredictive models that can operate effectively in the real world. To address\nthese limitations, we introduce a novel multi-modality dataset (MUVIM) that\ncontains four visual modalities: infra-red, depth, RGB and thermal cameras.\nThese modalities offer benefits such as obfuscated facial features and improved\nperformance in low-light conditions. We formulated fall detection as an anomaly\ndetection problem, in which a customized spatio-temporal convolutional\nautoencoder was trained only on ADLs so that a fall would increase the\nreconstruction error. Our results showed that infra-red cameras provided the\nhighest level of performance (AUC ROC=0.94), followed by thermal (AUC\nROC=0.87), depth (AUC ROC=0.86) and RGB (AUC ROC=0.83). This research provides\na unique opportunity to analyze the utility of camera modalities in detecting\nfalls in a home setting while balancing performance, passiveness, and privacy.", "title": "Multi Visual Modality Fall Detection Dataset", "url": "http://arxiv.org/abs/2206.12740v1" }

{ "text_length": 1831 }

{ "abstract": "This study evaluates the discriminating capacity (uniqueness) of the EEG data\nfrom the WAY EEG GAL public dataset to authenticate individuals against one\nanother as well as its permanence. In addition to the EEG data, Luciw et al.\nprovide EMG (Electromyography), and kinematics data for engineers and\nresearchers to utilize WAY EEG GAL for further studies. However, evaluating the\nEMG and kinematics data is outside the scope of this study. The goal of the\nstate-of-the-art is to determine whether EEG data can be utilized to control\nprosthetic devices. On the other hand, this study aims to evaluate the\nseparability of individuals through EEG data to perform user authentication. A\nfeature importance algorithm is utilized to select the best features for each\nuser to authenticate them against all others. The authentication platform\nimplemented for this study is based on Machine Learning models/classifiers. As\nan initial test, two pilot studies are performed using Linear Discriminant\nAnalysis (LDA) and Support Vector Machine (SVM) to observe the learning trends\nof the models by multi-labeling the EEG dataset. Utilizing kNN first as the\nclassifier for user authentication, accuracy around 75% is observed. Thereafter\nto improve the performance both linear and non-linear SVMs are used to perform\nclassification. The overall average accuracies of 85.18% and 86.92% are\nachieved using linear and non-linear SVMs respectively. In addition to\naccuracy, F1 scores are also calculated. The overall average F1 score of 87.51%\nand 88.94% are achieved for linear and non-linear SVMs respectively. Beyond the\noverall performance, high performing individuals with 95.3% accuracy (95.3% F1\nscore) using linear SVM and 97.4% accuracy (97.3% F1 score) using non-linear\nSVM are also observed.", "title": "Examining Uniqueness and Permanence of the WAY EEG GAL dataset toward User Authentication", "url": "http://arxiv.org/abs/2209.04802v1" }

{ "text_length": 1909 }

{ "abstract": "The recent surge in machine learning augmented turbulence modelling is a\npromising approach for addressing the limitations of Reynolds-averaged\nNavier-Stokes (RANS) models. This work presents the development of the first\nopen-source dataset, curated and structured for immediate use in machine\nlearning augmented turbulence closure modelling. The dataset features a variety\nof RANS simulations with matching direct numerical simulation (DNS) and\nlarge-eddy simulation (LES) data. Four turbulence models are selected to form\nthe initial dataset: $k$-$\\varepsilon$, $k$-$\\varepsilon$-$\\phi_t$-$f$,\n$k$-$\\omega$, and $k$-$\\omega$ SST. The dataset consists of 29 cases per\nturbulence model, for several parametrically sweeping reference DNS/LES cases:\nperiodic hills, square duct, parametric bumps, converging-diverging channel,\nand a curved backward-facing step. At each of the 895,640 points, various RANS\nfeatures with DNS/LES labels are available. The feature set includes quantities\nused in current state-of-the-art models, and additional fields which enable the\ngeneration of new feature sets. The dataset reduces effort required to train,\ntest, and benchmark new models. The dataset is available at\nhttps://doi.org/10.34740/kaggle/dsv/2044393 .", "title": "A curated dataset for data-driven turbulence modelling", "url": "http://arxiv.org/abs/2103.11515v1" }

{ "text_length": 1336 }

{ "abstract": "Following QUIC protocol ratification on May 2021, the third major version of\nthe Hypertext Transfer Protocol, namely HTTP/3, was published around one year\nlater in RFC 9114. In light of these consequential advancements, the current\nwork aspires to provide a full-blown coverage of the following issues, which to\nour knowledge have received feeble or no attention in the literature so far.\nFirst, we provide a complete review of attacks against HTTP/2, and elaborate on\nif and in which way they can be migrated to HTTP/3. Second, through the\ncreation of a testbed comprising the at present six most popular HTTP/3-enabled\nservers, we examine the effectiveness of a quartet of attacks, either stemming\ndirectly from the HTTP/2 relevant literature or being entirely new. This\nscrutiny led to the assignment of at least one CVE ID with a critical base\nscore by MITRE. No less important, by capitalizing on a realistic, abundant in\ndevices testbed, we compiled a voluminous, labeled corpus containing traces of\nten diverse attacks against HTTP and QUIC services. An initial evaluation of\nthe dataset mainly by means of machine learning techniques is included as well.\nGiven that the 30 GB dataset is made available in both pcap and CSV formats,\nforthcoming research can easily take advantage of any subset of features,\ncontingent upon the specific network topology and configuration.", "title": "A hands-on gaze on HTTP/3 security through the lens of HTTP/2 and a public dataset", "url": "http://arxiv.org/abs/2208.06722v2" }

{ "text_length": 1495 }

{ "abstract": "Quantum machine learning (QML) networks promise to have some computational\n(or quantum) advantage for classifying supervised datasets (e.g., satellite\nimages) over some conventional deep learning (DL) techniques due to their\nexpressive power via their local effective dimension. There are, however, two\nmain challenges regardless of the promised quantum advantage: 1) Currently\navailable quantum bits (qubits) are very small in number, while real-world\ndatasets are characterized by hundreds of high-dimensional elements (i.e.,\nfeatures). Additionally, there is not a single unified approach for embedding\nreal-world high-dimensional datasets in a limited number of qubits. 2) Some\nreal-world datasets are too small for training intricate QML networks. Hence,\nto tackle these two challenges for benchmarking and validating QML networks on\nreal-world, small, and high-dimensional datasets in one-go, we employ quantum\ntransfer learning composed of a multi-qubit QML network, and a very deep\nconvolutional network (a with VGG16 architecture) extracting informative\nfeatures from any small, high-dimensional dataset. We use real-amplitude and\nstrongly-entangling N-layer QML networks with and without data re-uploading\nlayers as a multi-qubit QML network, and evaluate their expressive power\nquantified by using their local effective dimension; the lower the local\neffective dimension of a QML network, the better its performance on unseen\ndata. Our numerical results show that the strongly-entangling N-layer QML\nnetwork has a lower local effective dimension than the real-amplitude QML\nnetwork and outperforms it on the hard-to-classify three-class labelling\nproblem. In addition, quantum transfer learning helps tackle the two challenges\nmentioned above for benchmarking and validating QML networks on real-world,\nsmall, and high-dimensional datasets.", "title": "Quantum Transfer Learning for Real-World, Small, and High-Dimensional Datasets", "url": "http://arxiv.org/abs/2209.07799v4" }

{ "text_length": 1964 }

{ "abstract": "Owing to the continuous efforts by the Chinese NLP community, more and more\nChinese machine reading comprehension datasets become available. To add\ndiversity in this area, in this paper, we propose a new task called Sentence\nCloze-style Machine Reading Comprehension (SC-MRC). The proposed task aims to\nfill the right candidate sentence into the passage that has several blanks. We\nbuilt a Chinese dataset called CMRC 2019 to evaluate the difficulty of the\nSC-MRC task. Moreover, to add more difficulties, we also made fake candidates\nthat are similar to the correct ones, which requires the machine to judge their\ncorrectness in the context. The proposed dataset contains over 100K blanks\n(questions) within over 10K passages, which was originated from Chinese\nnarrative stories. To evaluate the dataset, we implement several baseline\nsystems based on the pre-trained models, and the results show that the\nstate-of-the-art model still underperforms human performance by a large margin.\nWe release the dataset and baseline system to further facilitate our community.\nResources available through https://github.com/ymcui/cmrc2019", "title": "A Sentence Cloze Dataset for Chinese Machine Reading Comprehension", "url": "http://arxiv.org/abs/2004.03116v2" }

{ "text_length": 1229 }

{ "abstract": "The Machine Learning for Combinatorial Optimization (ML4CO) NeurIPS 2021\ncompetition aims to improve state-of-the-art combinatorial optimization solvers\nby replacing key heuristic components with machine learning models. On the dual\ntask, we design models to make branching decisions to promote the dual bound\nincrease faster. We propose a knowledge inheritance method to generalize\nknowledge of different models from the dataset aggregation process, named KIDA.\nOur improvement overcomes some defects of the baseline\ngraph-neural-networks-based methods. Further, we won the\n$1$\\textsuperscript{st} Place on the dual task. We hope this report can provide\nuseful experience for developers and researchers. The code is available at\nhttps://github.com/megvii-research/NeurIPS2021-ML4CO-KIDA.", "title": "ML4CO-KIDA: Knowledge Inheritance in Dataset Aggregation", "url": "http://arxiv.org/abs/2201.10328v3" }

{ "text_length": 879 }

{ "abstract": "Most eCommerce applications, like web-shops have millions of products. In\nthis context, the identification of similar products is a common sub-task,\nwhich can be utilized in the implementation of recommendation systems, product\nsearch engines and internal supply logistics. Providing this data set, our goal\nis to boost the evaluation of machine learning methods for the prediction of\nthe category of the retail products from tuples of images and descriptions.", "title": "A Retail Product Categorisation Dataset", "url": "http://arxiv.org/abs/2103.13864v2" }

{ "text_length": 534 }

{ "abstract": "The best neural architecture for a given machine learning problem depends on\nmany factors: not only the complexity and structure of the dataset, but also on\nresource constraints including latency, compute, energy consumption, etc.\nNeural architecture search (NAS) for tabular datasets is an important but\nunder-explored problem. Previous NAS algorithms designed for image search\nspaces incorporate resource constraints directly into the reinforcement\nlearning (RL) rewards. However, for NAS on tabular datasets, this protocol\noften discovers suboptimal architectures. This paper develops TabNAS, a new and\nmore effective approach to handle resource constraints in tabular NAS using an\nRL controller motivated by the idea of rejection sampling. TabNAS immediately\ndiscards any architecture that violates the resource constraints without\ntraining or learning from that architecture. TabNAS uses a Monte-Carlo-based\ncorrection to the RL policy gradient update to account for this extra filtering\nstep. Results on several tabular datasets demonstrate the superiority of TabNAS\nover previous reward-shaping methods: it finds better models that obey the\nconstraints.", "title": "TabNAS: Rejection Sampling for Neural Architecture Search on Tabular Datasets", "url": "http://arxiv.org/abs/2204.07615v4" }

{ "text_length": 1272 }

{ "abstract": "In machine learning, incorporating more data is often seen as a reliable\nstrategy for improving model performance; this work challenges that notion by\ndemonstrating that the addition of external datasets in many cases can hurt the\nresulting model's performance. In a large-scale empirical study across\ncombinations of four different open-source chest x-ray datasets and 9 different\nlabels, we demonstrate that in 43% of settings, a model trained on data from\ntwo hospitals has poorer worst group accuracy over both hospitals than a model\ntrained on just a single hospital's data. This surprising result occurs even\nthough the added hospital makes the training distribution more similar to the\ntest distribution. We explain that this phenomenon arises from the spurious\ncorrelation that emerges between the disease and hospital, due to\nhospital-specific image artifacts. We highlight the trade-off one encounters\nwhen training on multiple datasets, between the obvious benefit of additional\ndata and insidious cost of the introduced spurious correlation. In some cases,\nbalancing the dataset can remove the spurious correlation and improve\nperformance, but it is not always an effective strategy. We contextualize our\nresults within the literature on spurious correlations to help explain these\noutcomes. Our experiments underscore the importance of exercising caution when\nselecting training data for machine learning models, especially in settings\nwhere there is a risk of spurious correlations such as with medical imaging.\nThe risks outlined highlight the need for careful data selection and model\nevaluation in future research and practice.", "title": "When More is Less: Incorporating Additional Datasets Can Hurt Performance By Introducing Spurious Correlations", "url": "http://arxiv.org/abs/2308.04431v1" }

{ "text_length": 1789 }

{ "abstract": "Data-driven science is an emerging paradigm where scientific discoveries\ndepend on the execution of computational AI models against rich,\ndiscipline-specific datasets. With modern machine learning frameworks, anyone\ncan develop and execute computational models that reveal concepts hidden in the\ndata that could enable scientific applications. For important and widely used\ndatasets, computing the performance of every computational model that can run\nagainst a dataset is cost prohibitive in terms of cloud resources. Benchmarking\napproaches used in practice use representative datasets to infer performance\nwithout actually executing models. While practicable, these approaches limit\nextensive dataset profiling to a few datasets and introduce bias that favors\nmodels suited for representative datasets. As a result, each dataset's unique\ncharacteristics are left unexplored and subpar models are selected based on\ninference from generalized datasets. This necessitates a new paradigm that\nintroduces dataset profiling into the model selection process. To demonstrate\nthe need for dataset-specific profiling, we answer two questions:(1) Can\nscientific datasets significantly permute the rank order of computational\nmodels compared to widely used representative datasets? (2) If so, could\nlightweight model execution improve benchmarking accuracy? Taken together, the\nanswers to these questions lay the foundation for a new dataset-aware\nbenchmarking paradigm.", "title": "A Case for Dataset Specific Profiling", "url": "http://arxiv.org/abs/2208.03315v1" }

{ "text_length": 1533 }