Find research datasets worth reusing
Search datasets from major research repositories and use ShareScore to quickly assess how well each record supports discovery, access, and reuse.
921
datasets available to search
ShareScore release 0.7.1
Dataset results
921 results for “Neural Network”
Figure 9. Classification accuracy regardless the ethnic group (Total accuracy 75%)-Impact of Ethnic Group on Human Emotion Recognition Using Backpropagation Neural Network
<p>Our experiments show that, the impact of ethnic group on the accuracy of emotions<br> recognition is a positive where the accuracy of emotion recognition considering ethnic group is<br> 83.3% as shown in Figure 8, and we got 75% of accuracy regardless ethnic group as shown in<br> Figure 9.</p>
Figure 8. Classification accuracy of emotions considering the ethnic group-Impact of Ethnic Group on Human Emotion Recognition Using Backpropagation Neural Network
<p>To study the accuracy of emotion recognition for our approach regardless the ethnic group<br> we used 108 images for the training representing six emotions of six persons. For testing, we used<br> 36 images representing six emotions of six persons.<br> On the other hand, to study the accuracy of emotion recognition for our approach<br> considering the ethnic group we used 36 images for training for each ethnic group representing six<br> emotions of six persons, and test the classifier by using 12 images representing six emotions of six<br> persons.</p>
Train and test datasets used for the paper "Neural network time-series classifiers for gravitational-wave searches in single-detector periods"
<p>This repository contains the datasets used for training and testing during the work discussed in the paper "<a href="https://iopscience.iop.org/article/10.1088/1361-6382/ad40f0" target="_blank" rel="noopener">Neural network time-series classifiers for gravitational-wave searches in single-detector periods</a>". Please refer to this paper for more details on how the dataset was produced and cite it if you use these data:</p> <p><em>A. Trovato et al "Neural network time-series classifiers for gravitational-wave searches in single-detector periods", Class. Quant. Grav. 2024 DOI 10.1088/1361-6382/ad40f0.</em></p> <p>In this repository you will find six files in format npz, three of which refer to the test dataset and three to the train dataset. Each file name is of the type {label}_{train or test}.npz where "label" can be "glitch", "noise" or "signal", while the second part of the name indicates whether the file was used for training or testing.</p> <p>Each file is a collection of numpy arrays so it should be read with python. It contains 3 numpy arrays: 'X', 'Y' and 'metadata'. 'X' is a matrix containing 1-second segments of data sampled at 2048 Hz of the LIGO-Livingston detector, so it has shape: (number of samples, 2048). 'Y' contains the label for each segment, which is 0 for noise, 1 for signal and 2 for glitch, so it has shape: (number of samples,). In this case, the information on 'Y' is redundant since it's given directly by the filename. The 'metadata' matrix contains 17 metadata for each sample only for the case of signals, for glitch or noise it contains just 17 zeros for each sample. The shape of 'metadata' is thus: (number of samples, 17). For the signal files, for each sample the metadata is an array with these components:</p> <ol> <li>GPS start of the file from which this segment comes</li> <li>starting GPS time of this segment</li> <li>duration of the segment [s]</li> <li>mass1 [solar masses]</li> <li>mass2 [solar masses]</li> <li>spin1z</li> <li>spin2z</li> <li>inclination [radians]</li> <li>coalescence phase [radians]</li> <li>distance [Mpc]</li> <li>right_ascension [radians]</li> <li>declination [radians]</li> <li>polarization [radians]</li> <li>SNR (signal to noise ratio)</li> <li>shift of the signal w.r.t. the timeseries [s]</li> <li>length of the signal [s]</li> <li>fraction of the signal contained in the time window</li> </ol> <p>Number of samples:</p> <ul> <li>80000 for the file glitch_test.npz</li> <li>69998 for the file glitch_train.npz</li> <li>500000 for the file noise_test.npz</li> <li>250000 for the file noise_train.npz</li> <li>500000 for the file signal_test.npz</li> <li>250000 for the file signal_train.npz</li> </ul> <p>An example of few lines of python code to read each file is:</p> <pre><code>import numpy as np f = np.load("filename.npz") X = f['X'] Y = f['Y'] m = f['metadata'] </code></pre> <p>For the preparation of these data, we acknowledge the use of the following software packages: GWpy [1], PyCBC [2] and LALSuite [3]. </p> <p>This research has made use of data or software obtained from the Gravitational Wave Open Science Center (<a href="https://gwosc.org/" target="_blank" rel="noopener">gwosc.org</a>), a service of the LIGO Scientific Collaboration, the Virgo Collaboration, and KAGRA. This material is based upon work supported by NSF's LIGO Laboratory which is a major facility fully funded by the National Science Foundation, as well as the Science and Technology Facilities Council (STFC) of the United Kingdom, the Max-Planck-Society (MPS), and the State of Niedersachsen/Germany for support of the construction of Advanced LIGO and construction and operation of the GEO600 detector. Additional support for Advanced LIGO was provided by the Australian Research Council. Virgo is funded, through the European Gravitational Observatory (EGO), by the French Centre National de Recherche Scientifique (CNRS), the Italian Istituto Nazionale di Fisica Nucleare (INFN) and the Dutch Nikhef, with contributions by institutions from Belgium, Germany, Greece, Hungary, Ireland, Japan, Monaco, Poland, Portugal, Spain. KAGRA is supported by Ministry of Education, Culture, Sports, Science and Technology (MEXT), Japan Society for the Promotion of Science (JSPS) in Japan; National Research Foundation (NRF) and Ministry of Science and ICT (MSIT) in Korea; Academia Sinica (AS) and National Science and Technology Council (NSTC) in Taiwan.</p> <p>[1] https://gwpy.github.io<br>[2] https://pycbc.org<br>[3] https://lscsoft.docs.ligo.org/lalsuite</p>
Goat-CNN: A Lightweight Convolutional Neural Network for Pose-Independent Body Condition Score Estimation in Goats
<p>Here we introduce the dataset utilized in our published paper entitled "<a href="https://www.sciencedirect.com/science/article/pii/S2666154324002114">Goat-CNN: A Lightweight Convolutional Neural Network for Pose-Independent Body Condition Score Estimation in Goats</a>".</p> <p>Contained within the "bcs" folder are all the videos collected for this study. Each video file is named with a format denoting its respective details. The first number signifies the sequence of collection, the second denotes the ear tag, and the final figure represents the body condition score (BCS) value.</p> <p>For example: "1_158734_2.50" indicates the first sampling of an animal with the ear tag "158734" and a BCS value of "2.50".</p> <p>Additionally, we provide two Python scripts in this repository. The first script, "Video2Frame.py", facilitates the splitting of videos into individual frames. The second script, "Frames2npy.py", converts these frames into two numpy-friendly files with the extension ".npy". These files contain both the images ("X_train_bcs300.npy") and their corresponding labels ("Y_train_bcs300.npy").</p> <p>Furthermore, for the convenience of swift experimentation, we have included the desired .npy files within the repository.</p> <p>To load these files into your Python environment, you can use the following code snippet:</p> <div> <div>th4figs = '/content/drive/MyDrive/compag_2023/'</div> <br> <div>path4images = "/content/drive/MyDrive/CodeRefarm/datasets/BCS/X_train_bcs300.npy"</div> <div>Xtrain = np.load(path4images)</div> <br> <div>path4labels = "/content/drive/MyDrive/CodeRefarm/datasets/BCS/Y_train_bcs300.npy"</div> <div>Ytrain = np.load(path4labels).astype(float)</div> <br> <div>print("X train : ", Xtrain.shape)</div> <div>print("Y train : ", Ytrain.shape)</div> <div> <div> <div> <div> <div> <div> <div> </div> </div> <div> </div> </div> </div> </div> </div> <div> <div> <div> <div> <div> <div> <div> <div> <pre>X train : (5332, 300, 300, 3) Y train : (5332,)<br> </pre> </div> </div> </div> </div> </div> </div> </div> </div> </div> </div>
Physics-informed neural networks (PINNs) with unsaturated water flow models for inverse analysis of soil hydraulic parameters of layered soil profiles
<p>Information about the spatial distribution of soil hydraulic parameters is necessary for the accurate prediction of soil water flow and coupled movement of chemicals and heat at the field scale using a process-based model. Physics-informed neural networks (PINNs), which can provide physical constraints in deep learning to obtain a mesh-free solution, can be used to inversely estimate the soil hydraulic parameters from less and noisy training data. Previous studies using PINNs have successfully estimated soil hydraulic parameters for homogeneous soil but estimating such parameters of layered soil profiles where the interface depth and the parameters are unknown still has some difficulties. The objective of this study was to develop PINNs to inversely estimate the distribution of soil hydraulic parameters, such as saturated hydraulic conductivity and <em>α</em> and <em>n</em>, of the Mualem-van Genuchten model directly within layered soil profiles by predicting changes in pressure head from training data based on simulation results at given depths during infiltration. The impact of factors affecting PINNs performance, such as the weights assigned to each component of the loss function, the time range used in error computations, and the number of samples used to assess physical constraint was investigated. By assigning a larger weight to the physical constraint and excluding the earlier stage of infiltration in the loss function, the changes in pressure head and the three soil hydraulic parameter distributions within the layered soil profiles were successfully estimated. The developed PINNs can be further applied to more complex soils and can be improved.</p>
Conformer datasets for "Equivariant Graph Neural Networks for Toxicity Prediction"
<p>Predictive modeling of toxicity is a crucial step in the drug discovery pipeline. It can help filter out molecules with a high probability of failing in the early stages of de novo drug design. Thus, several machine learning (ML) models have been developed to predict the toxicity of molecules by combining classical ML techniques or deep neural networks with well-known molecular representations such as fingerprints or 2D graphs. But the more natural, accurate representation of molecules is expected to be defined in physical 3D space like in ab initio methods. Recent studies successfully used equivariant graph neural networks (EGNNs) for representation learning based on 3D structures to predict quantum-mechanical properties of molecules. Inspired by this, we investigated the performance of EGNNs to construct reliable ML models for toxicity prediction. We used the equivariant transformer (ET) model in TorchMD-NET for this. Eleven toxicity data sets taken from MoleculeNet, TDCommons, and ToxBenchmark have been considered to evaluate the capability of ET for toxicity prediction. Our results show that ET adequately learns 3D representations of molecules that can successfully correlate with toxicity activity, achieving good accuracies on most data sets comparable to state-of-the-art models. We also test a physicochemical property, namely, the total energy of a molecule, to inform the toxicity prediction with a physical prior. However, our work suggests that these two properties can not be related. We also provide an attention weight analysis for helping to understand the toxicity prediction in 3D space and thus increase the explainability of the ML model. In summary, our findings offer promising insights considering 3D geometry information via EGNNs and provide a straightforward way to integrate molecular conformers into ML-based pipelines for predicting and investigating toxicity prediction in physical space. We expect that in the future, especially for larger, more diverse data sets, EGNNs will be an essential tool in this domain.</p> <p>PAPER</p> <p>https://pubs.acs.org/doi/full/10.1021/acs.chemrestox.3c00032</p> <p>CODE and MODELS:</p> <p>The conformer data sets and trained toxicity models will be published upon acceptance of this work. The code has been made available at <a href="https://github.com/jule-c/ET-Tox">https://github.com/jule-c/ET-Tox</a>, and the processed data as well as pretrained models for training and testing can be downloaded from <a href="../record/7942946">https://zenodo.org/record/7942946</a>. We can provide the full list of conformers as XYZ files upon request.</p>
Dataset for training SENMAP, a automatic tool to curate LTR-retrotransposons using convolutional neural networks
<p>Transposable elements (TEs) are specific structures of the genome of species, which can move from one location to another. For that reason, they can cause mutations or changes that can be negative, such as the appearance of diseases, or beneficial, such as participating in fundamental roles in the evolution of genomes and genetic diversity. Long Terminal Repeat retrotransposons (LTR-RT) are the most abundant in plant species, hence the importance of studying these structures in particular. Over the time, these elements can suffer changes called nested insertions, which can inactivate or modify the functioning of the element, for that they are no longer consider as intact element and cannot be used for identification and classification studies. We create a dataset containing 56,442 LTR-RTs targed as "non-intact" elements and 49,215 considered as "intact". </p> <p>We formated the sequences IDs in order to keep relevant information as the superfamily and the lineage, as well as the category (Negative for "non-intact" and Positive for "intact" elements). </p> <p> This dataset (the npy files obtained from the fasta file) was used for training SENMAP, a convolutional neural network architecture to obtain intact LTR-RT sequences in plant genomes, which is composed by four convolutional layers, LeakyReLU as activation function and BinaryFocalLoss as loss function. Achieving an F1-score percentage of 91.37% with test data, identifying low quality sequences rapidly and efficiently, contributing to curate libraries of LTR retrotransposons of plants genomes published in large-scale sequencing projects due to the post-genomic era.</p>
Real-Time Adaptation of an Artificial Neural Network for Transfemoral Amputees Using a Powered Prosthesis
<p>This dataset contains the data used in our manuscript titled "Real-Time Adaptation of an Artificial Neural Network for Transfemoral Amputees Using a Powered Prosthesis". Data structure is explained in the README.txt file located at the top-level of the dataset.</p> <p>Please contact corresponding author Richard B. Woodward for any questions.</p>
Robust quantum dots charge autotuning using neural network uncertainty - Output data
<p>Outputs of the model training and the offline autotuning experiments presented in the paper: "<em>Robust quantum dots charge autotuning using neural network uncertainty</em>".</p> <p>For convenience, the results are splitted in several zipped files:</p> <ul> <li><strong>run_outputs_light.zip</strong>: contains only settings and results text files (sufficient for compiling result tables).</li> <li><strong>run_outputs_full_scan.zip</strong>: contains complete scan of the diagrams (for qualitative analyse)</li> <li><strong>run_outputs_part<N>.zip</strong>: contains all autotuning simulation output, grouped by seed (images and video output types might vary between seeds)</li> </ul> <p>Each folder in the zipped files represent a run that includes:</p> <ul> <li>log file</li> <li>plots / images</li> <li>run settings</li> <li>performance results</li> <li>pytorch model parameters</li> </ul> <p>See README.txt for more information about the file strucutre.</p>
Self Consistent Recurrent Neural Network for Path Dependent Deformation
<p>Data and Machine Learning codes for the paper:</p> <ul> <li>Title<strong> : Self Consistent Recurrent Neural Network for Path Dependent Deformation</strong></li> </ul> <p><strong>Abstract</strong> : Current neural network (NN) structures can learn patterns from data points with historical dependence. Specifically, in natural language processing (NLP), sequential learning has transitioned from recurrence-based architectures to transformer-based architectures. However, it is not known in advance which NN architectures will perform best on datasets containing deformation history due to mechanical loading. Thus, this study ascertains the appropriateness of 1D-convolutional, recurrent, and transformer-based architectures for predicting material failure based on the earlier states in the form of deformation history. Following this investigation, the crucial issues arising from the mathematical computation process of the best-performing NN architectures and the physical properties of the deformation paths are examined in detail. Additionally, we propose a novel and adaptable RNN approach to address the fundamental challenges of truncation and consistency related to obtaining estimations that are compatible with the natural physical properties of deformation paths. This study will serve as a foundation for localization estimation and pave the way for future endeavors to propose further solutions to encountered challenges.</p>
Results from Interpreting Cis-Regulatory Interactions from Large-Scale Deep Neural Networks for Genomics
Open the record for dataset details and reuse information.
Video-Audio Neural Network Ensemble For Comprehensive Screening Of Autism Spectrum Disorder in Young Children (Openpose ADOS Dataset)
<p>Here, we share a de-identify subsample of the data used in the <a href="https://journals.plos.org/plosone/article?id=10.1371/journal.pone.0308388">research article</a>, that will allow interested scientists to test the <a href="https://github.com/AutismBrainBehavior/Video-Neural-Network-ASD-screening">shared code</a>, as well as, develop alternatives for achieving better prediction accuracy. We have prepared a subsample of pose estimation videos for the first 10 minutes of ADOS examination videos for each of the 160 children including in the current study (80 ASD and 80 TD, 80 Training set and 80 Testing set).</p> <p>With this subset of the full dataset, our trained model achieved an accuracy of 68.75% over 80 videos (40 ASD & 40 TD) by training the Visual Geometry Group 16 Long short term memory recurrent neural network (VGG16 LSTM RNN) over 80 training videos (40 ASD & 40 TD) at 64 batch size and 120 epochs.</p>
Output Dataset from "Stellar Atmospheric Parameters From Gaia BP/RP Spectra using Uncertain Neural Networks"
<p>Output dataset of stellar parameters and associated uncertainties derived by the machine learning approach, as decribed in <a href="https://doi.org/10.1093/mnras/stae1303" target="_blank" rel="noopener"><em>"Stellar Atmospheric Parameters From Gaia BP/RP Spectra using Uncertain Neural Networks", Fallows & Sanders</em> (</a><em><a href="https://doi.org/10.1093/mnras/stae1303" target="_blank" rel="noopener">2024)</a>. </em></p> <p>Included parameters: [Fe/H] (fe_h), effective temperature (teff), surface gravity (logg), [C/Fe] (c_fe), [N/Fe] (n_fe), and [a/M] (a_m). Gaia DR3 source ids are included for matching purposes, alongside Gaia bp_rp_excess_factor and ruwe metrics for quality filtering.</p> <p><em>We include predictions for only Gaia objects with radial velocity measurements ('output_0.0_360.0_v2.txt'), and for all* Gaia objects with XP spectra ('gaiaFull_output_0.0_360.0.zip'). Note our full catalogue has a total size of 30.9GB once uncompressed.</em></p> <p> </p> <p><em>* Not truly all objects with XP spectra; stars with spurious or unreliable measurements in our requred inputs (Gaia, 2MASS, WISE) have been removed.</em></p>
BOLD5000 Additional ROIs and RDMs for neural network research
<p>Artificial neural networks (ANNs) are sensitive to perturbations and adversarial attacks. One hypothesized solution to adversarial robustness is to align manifolds in the embedded space of neural networks with biologically grounded manifolds. Recent state-of-the-art works that emphasize learning robust neural representations, rather than optimizing for a specific target task like classification, support the idea that researchers should investigate this hypothesis. While works have shown that fine-tuning ANNs to coincide with biological vision does increase robustness to both perturbations and adversarial attacks, these works have relied on proprietary datasets- the lack of publicly available biological benchmarks make it difficult to evaluate the efficacy of these claims. Here, we deliver a curated dataset consisting of biological representations of images taken from two commonly used computer vision datasets, ImageNet and COCO, that can be easily integrated into model training and evaluation. Specifically, we take a large functional magnetic resonance imaging (fMRI) dataset (BOLD5000), preprocess it into representational dissimilarity matrices (RDMs), and establish an infrastructure that anyone can use to train models with biologically grounded representations. Using this infrastructure, we investigate the representations of several popular neural networks and find that as networks have been optimized for tasks, their correspondence with biological fidelity has decreased. Additionally, we use a previously unexplored graph-based technique, Fiedler partitioning, to showcase the viability of the biological data, and the potential to extend these analyses by extending RDMs into Laplacian matrices. Overall, our findings demonstrate the potential of utilizing our new biological benchmark to effectively enhance the robustness of models.</p>
BRAIN Journal-Brain-Like Artificial Intelligence for Automation-Figure 12. Affinities and Differences of Neuro-Symbolic Networks in Comparison to Classical Neural Networks
<p>After having briefly illustrated the basic function principle of neuro-symbolic networks, this<br> section aims at reviewing their affinities and differences to standard neural networks like for<br> example multi-layer perceptrons (MLPs) [58]. A summary of these affinities and differences is<br> given in Figure 12. The affinities concern certain functions of individual nodes of the networks. In<br> both cases, weighted input information is summed up and an activation function is applied to this<br> sum. In both cases, the individual nodes are interconnected to form networks. Much larger than the<br> number of affinities between neuro-symbolic networks and neural network is however the number<br> of differences. The first difference consists in the application domain. Neuro-symbolic networks<br> have so far mainly been applied for complex, large-scale sensor data processing of multimodal data<br> – an application which can so far barely be handled by neural networks.</p>
Figure 6. Interface of FFE program-Impact of Ethnic Group on Human Emotion Recognition Using Backpropagation Neural Network
<p>Points, Features Extraction and save all input information for the classifier (Features, Ethnic group,<br> Gender and emotion). Figure 6 shows the interface of FFE program.</p>
Figure 7. Samples of MSFDE dataset-Impact of Ethnic Group on Human Emotion Recognition Using Backpropagation Neural Network
<p>We performed groups of experiments to study the impact of ethnic group (race) in the<br> accuracy of emotion recognition with three kinds of ethnic groups (Asian, Caucasian as African).<br> So we have three experiments, each experiment has a neural network as a classifier, and each neural<br> network has three layers where there are 16 neurons in the hidden layer except Asian network has<br> 17 neurons (the best result with 17 neurons for Asians).</p>
Figure 3. 46 points are selected on face elements to describe the emotions.-Impact of Ethnic Group on Human Emotion Recognition Using Backpropagation Neural Network
<p>The number of points and the position of points are not standardized, but it is depending on<br> the features that will be extracted, and used for the classifier. Many researches use various number<br> of points and positions based on their view about the feature to be considered [13] [18] [19]. Figure<br> 3 shows the points we used.</p>
Figure 1. A proposed approach-of Ethnic Group on Human Emotion Recognition Using Backpropagation Neural Network
<p>Our proposed approach uses the face expression to detect the emotions through five steps<br> that shows in Figure 1.</p>
Figure. 2. Examples of Angry from different races. (A,B) African. (C,D) Asian. (E,F) Caucasian.-Impact of Ethnic Group on Human Emotion Recognition Using Backpropagation Neural Network
<p>We chose 46 points which are distributed over human face image and use these points for<br> features extraction. The choice of these points is to determine the shape of each element of the face<br> (eyes, eyebrows and mouth), because the shape of these elements is changeable for each emotion,<br> but these changes are different for each race as shown in Figure 2.</p>
ScienceDex guides
Understand access before you commit
These curated guides explain access requirements, typical timelines, costs, and reuse considerations for widely used research datasets.
Allen Brain Atlas
Allen Brain Atlas is an Allen Institute collection of brain map atlases, datasets, APIs, and analysis tools covering mouse, human, and non-human primate brain resources.
Annotated Behaviour and Observability Dataset (ABODe)
ABODe is a University of Edinburgh DataShare dataset for behavior classification in group-housed mice using home-cage video, identities, bounding boxes, ground-plate positions, and annotator labels.
DANDI Archive for NWB datasets
DANDI is a BRAIN Initiative archive for publishing and sharing neurophysiology data, including electrophysiology, optophysiology, and behavioral data packaged as NWB and related standards.
International Brain Laboratory public data
The International Brain Laboratory public data releases expose standardized mouse decision-making experiments, including Neuropixels recordings, widefield calcium imaging, behavior, and session metadata accessed through the ONE API.
OpenNeuro
OpenNeuro is a free, open platform for sharing neuroimaging datasets, with public search, dataset pages, and download paths for web, S3, DataLad, and the OpenNeuro CLI.