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.
173
datasets available to search
ShareScore release 0.7.1
Dataset results
173 results for “convolution neural network”
Charting nanocluster structures via convolutional neural networks
<p>The repository contains a notebook for the training of the autoencoder for the RDFs for structural classification. The notebook describes the procedure going from RDFs calculation to clustering of the reduced space. In the folder are contained Au147 structures, together with the associated pretrained AE, the 3D chart and the different clustering performed varying mean shift bandwidth.</p> <p>Files:</p> <p>- ChartAu147.ipynb: notebook</p> <p>- Configurations: directory with the dataset divided according to the CNA classification of the structures, xyz format with no headers, every 147 lines is a single structure</p> <p>- Libraries: directory with functions imported in the notebook</p> <p>- Precomputed: directory with the precomputed outputs</p> <p> - rdfs.npy: preocmputed RDFs of the data stored in configurations, npy format to load with NumPy</p> <p> - labels.npy: CNA labels of the RDFs, npy format to load with NumPy</p> <p> - model_au147.pth: pretrained model for au147</p> <p> - scaler_au147.pkl: minmax scaler of the RDFs</p> <p> - chart_3d.dat: 3d space generated via the encoder on the au147 dataset</p> <p> - ae_reconstructions.npy: reconstructions of the rdfs of the model (model_au147.pth)</p> <p> - MSscanbw: pretrained mean shift clustering with different bandwidths, the file "clus_vs_bw.dat" reports the number of clusters associated to each bandwidth</p>
Training Images for "ImmuNet" Convolutional Neural Network
<p>This dataset contains all annotations and images for training the machine learning architecture presented in this manscript:</p> <p>Shabaz Sultan, Mark A. J. Gorris, Lieke L. van der Woude, Franka Buytenhuijs, Evgenia Martynova, Sandra van Wilpe, Kiek Verrijp, Carl G. Figdor, I. Jolanda M. de Vries, Johannes Textor:<br>ImmuNet: a segmentation-free machine learning pipeline for immune landscape phenotyping in tumors by multiplex imaging.<br>Biology Methods and Protocols 10(1), bpae094, 2025. doi: 10.1093/biomethods/bpae094</p> <p>The .tar.gz file contains several multichannel images stored as TIFF files, and arranged in a folder structure that is convenient for matching the files to the annotations provided in the .json.gz file. We also provide an .h5 file that contains the final trained network that was used to generate the figures in this manuscript.</p> <p>Further information on the data can be found in the manuscript cited above. Instructions on how to use the annotations and the code can be found on our GitHub page at: https://github.com/jtextor/immunet</p>
Data for paper "Convolutional neural network-based statistical post-processing of ensemble precipitation forecasts"
<p>The forecasts and observation datasets are used in the paper "Convolutional neural network-based statistical post-processing of ensemble precipitation forecasts". https://doi.org/10.1016/j.jhydrol.2021.127301</p> <p>The forecast data is a subset of the "ensemble for machine learning dataset (ENS4ML)" from ECMWF. </p> <p>The Python codes are stored in Github: https://github.com/wentao-bnu/LeNet_CSG_Precip</p>
Convolutional Neural Networks for Classifying Combinatorial Metamaterials
<p>This dataset contains the training and test data, as well as the trained neural networks as used for the paper 'Machine Learning of Implicit Combinatorial Rules in Mechanical Metamaterials', as published in Physical Review Letters.</p> <p>In this paper, a neural network is used to classify each <span class="math-tex">\(k \times k\)</span> unit cell design of metamaterial M1 and M2 into one of two classes (C or I). Additionally, the performance of the trained networks is analysed in detail. A more detailed description of the contents of the dataset follows below.</p> <p><strong>NeuralNetwork_train_and_test_data.zip</strong></p> <p>This file contains the train and test data used to train the Convolutional Neural Networks (CNNs) of the paper. Each unit cell size has its own file, and is saved in a zipped numpy file type (.npz). It contains data for metamaterial M1 ("smiley_cube"), and metamaterial M2 classification (i) ("prek_xy") and (ii) ("unimodal_vs_oligomodal_inc_stripmodes").</p> <p><strong>CNN_saves_kxk.zip</strong></p> <p>This file contains the parameter configurations of the CNNs trained on <span class="math-tex">\(k \times k\)</span> unit cells for metamaterial M2 classification (ii). Classification (i) is denoted by an additional M2ii in the file name. Metamaterial M1 is denoted by an extra M1 in the file name. Every hyperparameter (number of filters<em> nf,</em> number of hidden neurons<em> nh</em>, learning rate<em> lr</em>) combination is saved separately. The neural networks can be loaded using Google's TensorFlow package in Python, specifically using the 'tf.keras.models.load_model' function. </p>
Data from Automated plankton image analysis using convolutional neural networks
<p>Datasets and code from Luo et al., "Automated plankton image analysis using convolutional neural networks." Limnology and Oceanography Methods.</p> <p>Data include:</p> <p>1) 42,564 item training library, sorted in 108 classes,</p> <p>2) 42,548 item test set for filtering thresholds, sorted into 38 groups. These images are independent from the training library, and are used for setting the thresholds for post-classification filtering.<br> CSV file: Luo_etal_FT_images_pred.csv contains the image name, predicted class, predicted probability, and validated group. Note that the file class_to_group.csv is needed to match up the class names to the group names.</p> <p>3) 75,000 item fully random, validated set for confusion matrix calculations, sorted into 38 groups. This set is a representation of the full dataset, selected at random after classification. <br> CSV file: Luo_etal_confusionmatrix_images.csv contains the image name, predicted class, predicted probability, and validated group. Note that the file class_to_group.csv is needed to match up the class names to the group names.</p> <p> </p> <p>Scripts and programs:</p> <p>1) Segmentation.zip contains the scripts and executables for the segmentation program.</p> <p>2) Plankton_template.zip contains the archived version of the SparseConvNet program used in manuscript (current version available at: https://github.com/btgraham/SparseConvNet or https://github.com/facebookresearch/SparseConvNet)<br> Note that google-sparsehash is necessary for running SparseConvNet.<br> Also, plankton_epoch-150.cnn are the weights from the training used in the manuscript, and should be placed in the /weights folder if you want to replicate the classifications.</p>
Trained convolutional neural network for the identification of long-duration mixed precipitation in Montréal (Canada)
<p>In this dataset the trained convolutional neural network is published that accompanies the paper "A deep learning approach for the identification of long-duration mixed precipitation in Montréal (Canada)" submitted to the special issue on "Machine-Learning Applications in the Atmospheric and Oceanic Sciences" by the journal Atmosphere&Ocean.</p> <p>The files were created using tensorflow in python. The trained network is available in .h5-format the history as numpy-file (npy).</p>
Robustness assessment of a C++ implementation of a quantized (int8) version of the LeNet-5 convolutional neural network
<p>The architecture of the LeNet-5 convolutional neural network (CNN) was defined by LeCun in its paper "Gradient-based learning applied to document recognition" (<a href="https://ieeexplore.ieee.org/document/726791">https://ieeexplore.ieee.org/document/726791</a>) to classify images of hand written digits (MNIST dataset).</p><p>This architecture has been customized to use Rectified Linear Unit (ReLU) as activation functions instead of Sigmoid, and 8-bit integers for weights and activations instead of floating-point.</p><p>It consists of the following layers:</p><ul><li><strong>conv1</strong>: Convolution 2D, 1 input channel (28x28), 3 output channels (28x28), kernel size 5, stride 1, padding 2.</li><li><strong>relu1</strong>: Rectified Linear Unit (3@28x28).</li><li><strong>max1</strong>: Subsampling buy max pooling (3@14x14).</li><li><strong>conv2</strong>: Convolution 2D, 3 input channels (14x14), 6 output channels (14x14), kernel size 5, stride 1, padding 2.</li><li><i><strong>relu2</strong></i>: Rectified Linear Unit (6@14x14).</li><li>max2: Subsampling buy max pooling (6@7x7).</li><li><i><strong>fc1</strong></i>: Fully connected (294, 147)</li><li><i><strong>fc2</strong></i>: Fully connected (147, 10)</li></ul><p>The fault hypotheses for this work include the occurrence of:</p><ul><li><strong>BF</strong>: single, double-adjacent and triple-adjacent bit-flip faults</li><li><strong>S0</strong>: single, double-adjacent and triple-adjacent stuck-at-0 faults</li><li><strong>S1</strong>: single, double-adjacent and triple-adjacent stuck-at-1 faults</li></ul><p>In the memory cells containing all the parameters of the CNN: </p><ul><li><strong>w</strong>: weights (int8)</li><li><strong>zw</strong>: zero point of the weights (int8)</li><li><strong>b</strong>: biases (int32)</li><li><strong>z</strong>: zero point (int8)</li><li><strong>m</strong>: m (int32)</li></ul><p>Images 200 to 249 from the MNIST dataset have been used as workload.</p><p>This dataset contains the raw data obtained from running exhaustive fault injection campaigns for all considered fault models, targeting all considered locations and for all the images in the workload.</p><p>In addition, the raw data have been lightly processed to obtain global data related to the particular bits and parameters affected by the faults, and the obtained failure modes.</p><h3>Files information</h3><ul><li><i>golden_run.csv</i>: Prediction obtained for all the images considered in the workload in the absence of faults (Golden Run). This is intended to act as oracle to determine the impact of injected faults. </li><li><i>single_faults/bit_flip</i> folder: Prediction obtained for all the images considered in the workload in presence of single bit-flip faults. There is one file for each parameter of each layer.</li><li><i>single_faults/stuck_at_0</i> folder: Prediction obtained for all the images considered in the workload in presence of single stuck-at-0 faults. There is one file for each parameter of each layer.</li><li><i>single_faults/stuck_at_1</i> folder: Prediction obtained for all the images considered in the workload in presence of single stuck-at-1 faults. There is one file for each parameter of each layer.</li><li><i>double_adjacent_faults/bit_flip</i> folder: Prediction obtained for all the images considered in the workload in presence of double adjacent bit-flip faults. There is one file for each parameter of each layer.</li><li><i>double_adjacent_faults/stuck_at_0</i> folder: Prediction obtained for all the images considered in the workload in presence of double adjacent stuck-at-0 faults. There is one file for each parameter of each layer.</li><li><i>double_adjacent_faults/stuck_at_1</i> folder: Prediction obtained for all the images considered in the workload in presence of double adjacent stuck-at-1 faults. There is one file for each parameter of each layer.</li><li><i>triple_adjacent_faults/bit_flip</i> folder: Prediction obtained for all the images considered in the workload in presence of triple adjacent bit-flip faults. There is one file for each parameter of each layer.</li><li><i>triple_adjacent_faults/stuck_at_0</i> folder: Prediction obtained for all the images considered in the workload in presence of triple adjacent stuck-at-0 faults. There is one file for each parameter of each layer.</li><li><i>triple_adjacent_faults/stuck_at_1</i> folder: Prediction obtained for all the images considered in the workload in presence of triple adjacent stuck-at-1 faults. There is one file for each parameter of each layer.</li></ul><h3>Methodology information</h3><p>First, the CNN was used to classify all the images of the workload in the absence of faults to get a reference to determine the impact of faults. This is golden_run.csv file.</p><p>After that, one fault injection experiment was executed for each bit of each element of each parameter of the CNN.</p><p>Each experiment consisted in:</p><ul><li>Affecting the bits (inverting it in case of bit-flip faults, setting it to 0 or 1 in case of stuck-at-0 or atuck-at-1 faults) identified by the mask.</li><li>Classifying all the images of the workload in the presence of this fault. The obtained output was stored in a given .csv file.</li><li>Removing the fault from the CNN by restoring the affected bits to its previous value.</li></ul><h3>List of variables (Name : Description (Possible values))</h3><ul><li><strong>IMGID</strong>: Integer number identifying the considered image (200-249).</li><li><strong>TENSORID</strong>: Integer number identiying the parameter affected by the fault (0 - No fault, 1 - conv1.w, 2 - conv1.zw, 3 - conv1.m, 4 - conv1.b, 5 - conv1.z, 6 - conv2.w, 7 - conv2.zw, 8 - conv2.m, 9 - conv2.b, 10 - conv2.z, 11 - fc1.w, 12 - fc1.zw, 13 - fc1.m, 14 - fc.b, 15 - fc1.z, 16 - fc2.w, 17 - fc2.zw, 18 - fc2.m, 19 - fc2.b, 20 - fc2.z)</li><li><strong>ELEMID</strong>: Integer number identiying the element of the parameter affected by the fault (-1 - No fault, [0-2] - {conv1.b, conv1.m, conv1.zw}, [0-74] - conv1.w, 0 - conv1.z, [0-5] - {conv2.b, conv2.m, conv2.zw}, [0-149] - conv2.w, 0 - {conv1.z, conv2.z, fc1.z, fc2.z}, [0-146] - {fc1.b, fc1.m, fc1.zw}, [0-43217] - fc1.w, [0-9] - {fc2.b, fc2.m, fc2.zw}, [0-1469] - fc2.w)</li><li><strong>MASK</strong>: 8-digit hexadecimal number identifying those bits affected by the fault ([00000000 - No fault, FFFFFFFF - all 32 bits faulty])</li><li><strong>FAULT</strong>: String identiying the type of fault (NF - No fault, BF - bit-flip, S0 - Stuck-at-0, S1 - Stuck-at-1)</li><li><strong>OUTPUT</strong>: 10 integer numbers provided by the CNN as output after processing the image. The highest value identifies the selected category for classification.</li><li><strong>SOFTMAX</strong>: 10 decimal numbers obtained after applying the softmax function to the provided output. They represent the probability of the image of belonging to the corresponding category for classification.</li><li><strong>PRED</strong>: Integer number representing the category predicted for the processed image.</li><li><strong>LABEL</strong>: integer number representing the actual category for the processed image.</li></ul>
Data for: Tang et al., Interpretable classification of Alzheimer's disease pathologies with a convolutional neural network pipeline. bioRxiv 2018.
<p>Datasets containing 63 whole slide images (WSIs) and their segmented 256x256 pixel tiles with approximately 80,000 tile-level amyloid-β pathology expert annotations.</p> <p><strong>Paper</strong>: "Interpretable classification of Alzheimer's disease pathologies with a convolutional neural network pipeline", bioRxiv 454793; DOI: <a href="https://doi.org/10.1101/454793">https://doi.org/10.1101/454793</a>.</p> <p><strong>Details:</strong> A total of 63 WSIs for 63 unique decedent cases spanning Alzheimer’s disease (AD) to non-AD and possessing a variety of CERAD scores. WSIs comprise three datasets as follows:</p> <ol> <li><em>Development (Phases I-II)</em>. 33 WSIs used for convolutional neural network (CNN) model development (29 training, 4 validation).</li> <li><em>Hold-out (Phase III)</em>. 10 WSIs selected by an expert neuropathologist as a held-out test set to assess the generalizability of the CNN model.</li> <li><em>CERAD-like hold-out</em>. 20 blinded WSIs collected solely for use in a CERAD-like scoring comparison study.</li> </ol> <p>Datasets 1 and 2 were color-normalized and segmented to 256x256 pixel image tiles for model training set (61,370 images), validation set (8,630 images), and hold-out test set (10,873 images). Dataset 3 was color-normalized but not segmented.</p> <p>Expert labels of plaques for Dataset 1 and 2 tiles are included in corresponding CSV files.</p> <p><strong>Slide source and preparation:</strong> All samples were retrieved from archives of the University of California, Davis Alzheimer’s Disease Center Brain Bank (<a href="https://www.ucdmc.ucdavis.edu/alzheimers/">https://www.ucdmc.ucdavis.edu/alzheimers/</a>). Archival samples analyzed in this study were 5 μm formalin fixed, paraffin embedded sections of the superior and middle temporal gyrus from human brain. The tissue had been previously stained with an amyloid-β antibody (4G8, recognizing residues 17-24, BioLegend, formerly Covance) that were first pretreated with formic acid to rid samples of endogenous protein. All slides were digitized using an Aperio AT2 up to 40x magnification.</p> <p><strong>Code:</strong> Please visit <a href="https://github.com/keiserlab/plaquebox-paper">https://github.com/keiserlab/plaquebox-paper</a></p> <p> </p>
Convolutional neural network for automated surface crack detection using inductive thermography
<p>Two phase images of the samples AIT_01 and AIT_08, analysed in the publication "Convolutional neural network for automated surface crack detection using inductive thermography", submitted to the Journal of Electronic Imaging.</p>
CNNpredIM - Dataset for Rapid Prediction of Earthquake Ground Shaking Intensity Using Raw Waveform Data and a Convolutional Neural Network
<p>The <strong>dataset</strong> available here is the dataset used in the <a href="https://academic.oup.com/gji/advance-article/doi/10.1093/gji/ggaa233/5836721"><strong>paper</strong> <em>"Rapid Prediction of Earthquake Ground Shaking Intensity Using Raw Waveform Data and a Convolutional Neural Network".</em></a></p> <p>The <strong>abstract</strong> of the <strong>paper</strong>:</p> <blockquote> <p>This study describes a deep convolutional neural network (CNN) based technique for the prediction of intensity measurements (IMs) of ground shaking. The input data to the CNN model consists of multistation 3C broadband and accelerometric waveforms recorded during the 2016 Central Italy earthquake sequence for M ≥ 3.0. We find that the CNN is capable of predicting accurately the IMs at stations far from the epicenter and that have not yet recorded the maximum ground shaking when using a 10 s window starting at the earthquake origin time. The CNN IM predictions do not require previous knowledge of the earthquake source (location and magnitude). Comparison between the CNN model predictions and the predictions obtained with Bindi et al. (2011) GMPE (which require location and magnitude) has shown that the CNN model features similar error variance but smaller bias. Although the technique is not strictly designed for earthquake early warning, we found that it can provide useful estimates of ground motions within 15-20 sec after earthquake origin time depending on various setup elements (e.g., times for data transmission, computation, latencies). The technique has been tested on raw data without any initial data pre-selection in order to closely replicate real-time data streaming. When noise examples were included with the earthquake data, the CNN was found to be stable predicting accurately the ground shaking intensity corresponding to the noise amplitude.</p> </blockquote>
Generalized linear model with elastic net regularization and convolutional neural network for evaluating Aphanomyces root rot severity in lentil
<p>Red-Green-Blue (RGB) imaging was used to evaluate Aphanomyces root rot in 547 lentil accessions and lines. The root images were pre-processed by removing image background. This dataset (6,460 root images) was used to build two machine learning models — generalized linear model with elastic net regularization and convolutional neural network— to classify root images into three classes. Details about the methodology and results are described in Marzougui et al. (2020, Plant Phenomics).</p> <p>The excel file includes Aphanomyces root rot disease visual scores (<em>Root_Rating</em>), unique identifier for each lentil accession/line (<em>Lentil_ID</em>), unique identifier for each experiment (<em>Experiment</em>), and unique identifier for each image (<em>Lab_ID</em>).</p>
Semantic Segmentation of Time Series Imagery Using Deep Convolutional Neural Networks: A Case Study of Sandbars in Grand Canyon
<p>This dataset contains imagery used to train and test Deep Convolutional Neural Networks for the purpose of binary semantic segmentation of a time series of oblique imagery capturing sandbar monitoring sites in The Grand Canyon. In addition the scripts needed for removing image distortion, registering, rectifying, and labeling imagery is present. </p>
Predicting placenta transcriptional regulatory interactions based on spatial gene expression data and convolutional neural network
<p><strong>Aims:</strong> The dysfunction of placenta development is correlated to the defects of pregnancy and fetal growth. The detailed molecular mechanism of placenta development is not identified in human due to the lack of material in vivo. Image-based reconstructions of GRN are still very underdeveloped.</p> <p><strong>Methods and Results:</strong> In this study, first-trimester chorionic villus and decidua tissues were collected. Next, we present a machine-learning system to infer gene interaction networks of the human placenta from immunofluorescence images of trophoblast specific transcription factors obtained by a high-resolution scanner.</p> <p><strong>Conclusions:</strong> The experimental results show that deep learning models reveal regulatory roles that have not yet been fully recognized. The spatial expression data reveal new regulatory relationships that traditional experiments have failed to recognize, and has allowed the development of gene regulation networks based on the spatial distribution of gene expression. We demonstrate the effectiveness of this approach in building networks using high-resolution images of the human placenta. Our analysis is of certain significance for further exploration of the development of the placenta and the occurrence of pregnancy-related diseases in the future. The datasets and analysis provide a useful source for the researchers in the field of the maternal-fetal interface and the establishment of pregnancy.</p>
Determining non-significant bits on a C++ implementation of the LeNet-5 convolutional neural network to be used for storing error correcting codes to protect weights and biases. Robustness assessment of the network after integrating the proposed codes.
<p>The architecture of the LeNet-5 convolutional neural network (CNN) was defined by LeCun in its paper "Gradient-based learning applied to document recognition" (<a href="https://ieeexplore.ieee.org/document/726791">https://ieeexplore.ieee.org/document/726791</a>) to classify images of hand written digits (MNIST dataset).</p><p>This architecture has been customized to use Rectified Linear Unit (ReLU) as activation functions instead of Sigmoid.</p><p>It consists of the following layers:</p><ul><li><strong>conv1</strong>: Convolution 2D, 1 input channel (28x28), 3 output channels (28x28), kernel size 5, stride 1, padding 2.</li><li><strong>relu1</strong>: Rectified Linear Unit (3@28x28).</li><li><strong>max1</strong>: Subsampling buy max pooling (3@14x14).</li><li><strong>conv2</strong>: Convolution 2D, 3 input channels (14x14), 6 output channels (14x14), kernel size 5, stride 1, padding 2.</li><li><strong>relu2</strong>: Rectified Linear Unit (6@14x14).</li><li><strong>max2</strong>: Subsampling buy max pooling (6@7x7).</li><li><strong>fc1</strong>: Fully connected (294, 147)</li><li><strong>fc2</strong>: Fully connected (147, 10)</li></ul><p>The fault hypotheses for this work include the occurrence of:</p><ul><li><strong>S0</strong>/<strong>S1</strong>: multiple adjacent stuck-at-0 and stuck-at-1 faults to determine the least significant bits of weights and biases that could be used to store the proposed error correcting codes.</li><li><strong>BF</strong>: single, double, and triple bit-flip faults to assess the robustness of the considered CNN</li></ul><p>In the memory cells containing all the parameters of the CNN: </p><ul><li><strong>w</strong>: weights (float32)</li><li><strong>b</strong>: biases (float32)</li></ul><p>All the images (10000) from the MNIST dataset have been used as workload.</p><p>The weights and biases of the LeNet-5 architecture have been protected using six different error correcting codes that have been deployed in the least significant bits of these elements.</p><p>The parity check matrices (H = P I) that define these ECCs are:</p><ul><li><strong>SEC(32, 26)</strong> (Hamming) under a <i>classic policy </i>(see methodology below):</li></ul><p><i> 11010010001000011101101000 100000</i></p><p><i> 10101001000100011011010100 010000</i></p><p><i> 01100100100010010110110010 001000</i></p><p><i> 00011100010001001110001101 000100</i></p><p><i> 00000011110000100001111011 000010</i></p><p><i> 00000000001111100000000111 000001</i></p><ul><li><strong>SEC(23, 18)</strong> (Hamming) under a <i>conservative policy</i> (see methodology below):</li></ul><p><i> 111100001111000000 10000</i></p><p><i> 110011101000111000 01000</i></p><p><i> 101011010100100110 00100</i></p><p><i> 010110110010010101 00010</i></p><p><i> 001101110001001011 00001</i></p><ul><li><strong>SEC(13, 9)</strong> (Hamming) under an <i>aggressive policy </i>(see methodology below):</li></ul><p><i> 110111000 1000</i></p><p><i> 101100110 0100</i></p><p><i> 011010101 0010</i></p><p><i> 111001011 0001</i></p><ul><li><strong>DEC(32, 21)</strong> (low redundancy and reduced overhead DEC) under a <i>classic policy </i>(see methodology below):</li></ul><p><i> 111000011001010010000 10000000000</i></p><p><i> 110110000011101000000 01000000000</i></p><p><i> 101011000110000010001 00100000000</i></p><p><i> 100101101000110001000 00010000000</i></p><p><i> 011010101100100000100 00001000000</i></p><p><i> 010101010100001001010 00000100000</i></p><p><i> 001100110010010100100 00000010000</i></p><p><i> 000011110001000110010 00000001000</i></p><p><i> 000000001111001101001 00000000100</i></p><p><i> 000000000000111100111 00000000010</i></p><p><i> 000000000000000011111 00000000001</i></p><ul><li><strong>DEC(28, 18)</strong> (low redundancy and reduced overhead DEC) under a <i>conservative policy </i>(see methodology below):</li></ul><p><i> 111111000000000000 1000000000</i></p><p><i> 110100111100000000 0100000000</i></p><p><i> 110000100011110000 0010000000</i></p><p><i> 001110010011001100 0001000000</i></p><p><i> 101100001010101010 0000100000</i></p><p><i> 010001001101010110 0000010000</i></p><p><i> 001011000101101001 0000001000</i></p><p><i> 101000011000110101 0000000100</i></p><p><i> 010001110000011011 0000000010</i></p><p><i> 000010100110000111 0000000001</i></p><ul><li><strong>DEC(17, 9)</strong> (low redundancy and reduced overhead DEC) under an <i>aggressive policy </i>(see methodology below):</li></ul><p><i> 111110000 10000000</i></p><p><i> 111001100 01000000</i></p><p><i> 110101010 00100000</i></p><p><i> 101010110 00010000</i></p><p><i> 101101001 00001000</i></p><p><i> 100110101 00000100</i></p><p><i> 100011011 00000010</i></p><p><i> 110000111 00000001</i></p><p>This dataset contains the raw data obtained from:</p><ul><li>running exhaustive fault injection campaigns for increasingly multiple stuck-at faults in the least significant bits of all weights and biases (simultaneously) and for all the images in the workload.</li><li>running statistical fault injection campaigns for single, double, and triple bit-flip faults, randomly targeting the considered locations and images in the workload.</li></ul><h3>Files information</h3><ul><li><i>no_ecc </i>folder: Results obtained for the original (not protected) version of the CNN.<ul><li><i>golden_run.csv</i>: Prediction obtained for all the images considered in the workload in the absence of faults (Golden Run). This is intended to act as oracle to determine the impact of injected faults.</li><li><i>sampling_SBF_10000.csv</i>: Prediction obtained for running 10000 statistical fault injection experiments for single bit-flip faults.</li><li><i>sampling_DBF_10000.csv</i>: Prediction obtained for running 10000 statistical fault injection experiments for double bit-flip faults.</li><li><i>sampling_TBF_10000.csv</i>: Prediction obtained for running 10000 statistical fault injection experiments for triple bit-flip faults.</li><li><i>locating_sensitive_bits </i>folder: Prediction obtained for all the images considered in the workload in presence of stuck-at-0/stuck-at-1 faults that simultaneously target the N least significant bits of all weights and biases. There is one file for each parameter of type of fault and range of targeted bits. Files for bits in the range [11, 0] are not included as they obtain eactly the same results as the Golden Run (faults do not alter the behaviour of the network).</li></ul></li><li><i>sec/classic</i>, <i>sec/conservative</i>, and <i>sec/aggressive</i> folders: They contain the results obtained for the CNN protected by SEC(32, 26), SEC(23, 18), and SEC(13, 9), respectively.<ul><li><i>golden_run.csv</i>: Prediction obtained for all the images considered in the workload in the absence of faults (Golden Run). This is intended to act as oracle to determine the impact of injected faults. It must be noted that this file could be different that the golden_run.csv file for the original version of the CNN, as deploying the ECC in the weights and biases may have affected the behaviour of the network.</li><li><i>sampling_SBF_10000.csv</i>: Prediction obtained for running 10000 statistical fault injection experiments for single bit-flip faults. They should all be tolerated by the definition of the ECC.</li><li><i>sampling_DBF_10000.csv</i>: Prediction obtained for running 10000 statistical fault injection experiments for double bit-flip faults. They could be more harmful than for the unprotected version of the CNN, as the ECC may erroneously flip correct bits.</li></ul></li><li><i>dec/classic</i>, <i>dec/conservative</i>, and <i>dec/aggressive </i>folders: They contain the results obtained for the CNN protected by DEC(32, 21), DEC(28, 18), and DEC(17, 9), respectively.<ul><li><i>golden_run.csv</i>: Prediction obtained for all the images considered in the workload in the absence of faults (Golden Run). This is intended to act as oracle to determine the impact of injected faults. It must be noted that this file could be different that the golden_run.csv file for the original version of the CNN, as deploying the ECC in the weights and biases may have affected the behaviour of the network.</li><li><i>sampling_DBF_10000.csv</i>: Prediction obtained for running 10000 statistical fault injection experiments for double bit-flip faults. They should all be tolerated by the definition of the ECC.</li><li><i>sampling_TBF_10000.csv</i>: Prediction obtained for running 10000 statistical fault injection experiments for triple bit-flip faults. They could be more harmful than for the unprotected version of the CNN, as the ECC may erroneously flip correct bits.</li></ul></li></ul><h3>Methodology information</h3><p>First, the CNN was used to classify all the images of the workload in the absence of faults to get a reference to determine the impact of faults. This is <i>golden_run.csv</i> file.</p><p>To locate non-significant bits in weights and biases, fault injection experiments were executed targeting all elements of all parameters of the CNN using the following procedure:</p><ul><li>The initial mask targeted only the least significant bit</li><li>Until the mask targets all bits of the elements (32 bits as they are single-precision floating point values):<ul><li>Affect the bits (setting them to 0 or 1 in case of stuck-at-0 or stuck-at-1 faults) identified by the mask for all elements of all parameters.</li><li>Classify all the images of the workload in the presence of this fault. The obtained output was stored in a given .csv file.</li><li>Remove the fault from the CNN by restoring the affected bits to its previous value.</li><li>Add the next adjacent bit to the mask, so it targets an additional least significant bit.</li></ul></li></ul><p>The analysis of the obtained results may help in determining which bits can be used to store an ECC:</p><ul><li>which bits never affect the behaviour of the CNN, as the predicted classification is exactly the same than in the absence of faults.</li><li>which bits midly affect the behaviour of the CNN, as although the predicted classifications differ from those in the absence of faults, the accuracy of the network is barely affected.</li><li>which bits greatly affect the behaviour of the CNN, as the accuracy of the network is significantly affected.</li></ul><p>Accordingly, three different policies have been identified for deploying an ECC using these bits:</p><ul><li><strong>Classic policy</strong>: The ECC protects as much bits as possible.</li><li><strong>Conservative policy</strong>: The ECC protects all those bits that may affect the prediction of the network.</li><li><strong>Aggressive policy</strong>: The ECC protects only those bits that significantly affect the accuracy of the network.</li></ul><p>After designing and deploying a single ECC and a double ECC for each of the identified policies, fault injection experiments were executed to verify their behaviour in the presence of faults.</p><p>Single and double ECCs were tested against single and double bit-flip, respectively (all faults should be tolerated,) and double and triple bit-flips, respectively (a correct bit could be erroneously flipped.)</p><p>Due to the heavy computational load of the decoders, statistical injection was used to run the required fault injection campaigns with a sample size (number of experiments) of 10000.</p><p>Each experiment consisted in:</p><ul><li>Randomly selecting the image to process, and the parameter, element, and bits (mask) to be targeted by the fault.</li><li>Affecting the bits (inverting them) identified by the mask.</li><li>Classifying the selected image of the workload in the presence of this fault. The obtained output was stored in a given .csv file.</li><li>Removing the fault from the CNN by restoring the affected bits to its previous value.</li></ul><h3>List of variables (Name : Description (Possible values))</h3><ul><li><strong>IMGID</strong>: Integer number identifying the considered image (1-9999).</li><li><strong>TENSORID</strong>: Integer number identiying the parameter affected by the fault (0 - No fault, 1 - conv1.w, 2 - conv1.b, 3 - conv2.w, 4 - conv2.b, 5 - fc1.w, 6 - fc1.b, 7 - fc2.w, 8 - fc2.b).</li><li><strong>ELEMID</strong>: Integer number identiying the element of the parameter affected by the fault (-1 - No fault, [0-2] - conv1.b, [0-74] - conv1.w, [0-5] - conv2.b, [0-149] - conv2.w, [0-146] - fc1.b, [0-43217] - fc1.w, [0-9] - fc2.b, [0-1469] - fc2.w).</li><li><strong>MASK</strong>: 8-digit hexadecimal number identifying those bits affected by the fault ([00000000 - No fault, FFFFFFFF - all 32 bits faulty]).</li><li><strong>FAULT</strong>: String identiying the type of fault (NF - No fault, BF - bit-flip, S0 - Stuck-at-0, S1 - Stuck-at-1).</li><li><strong>SOFTMAX</strong>: 10 decimal numbers obtained after applying the softmax function to the provided output. They represent the probability of the image of belonging to the corresponding category for classification.</li><li><strong>PRED</strong>: Integer number representing the category predicted for the processed image.</li><li><strong>LABEL</strong>: integer number representing the actual category for the processed image.</li></ul>
Train and Evaluation Code, Road Classification Models and Test set of the paper "Impact of Image Resolution and Image Overlap on the Prediction Performance of Convolutional Neural Networks Trained for Road Classification"
<p>This repository contains the Python scripts built for training and evaluation of the implementation, together with the test data and the resulting road classification models corresponding to the paper "Impact of Image Resolution and Image Overlap on the Prediction Performance of Convolutional Neural Networks Trained for Road Classification". The scripts make use of the Tensorflow with Keras framework and the additional required dependencies.</p> <p>The training and validation set is based on the binary SROADEX dataset (https://zenodo.org/records/6482346) that was re-split into tiles that feature the image resolutions (256 x 256, 512 x 512, and 1024 x 1024 pixels) and image overlaps (0% and 12.5%) considered in this study. The data have been generated using scripts developed in Python using Open Source libraries (GDAL/OGR and MapScript) for rasterization of vector cartography that represents the axes of the different types of roads (urban, interurban and rural). This binary road data contains information from 16 full orthoimages (28.5 km * 18.5 km) with spatial resolution of 0.5 m/pixel from the insular and peninsular Spanish territory. Due to the size on disk of approximately 546 gigabytes, this training and validation data is only available upon request from the corresponding author. The test set has been generated from a novel area of 28.5 km * 18.5 km and features binary road labels. The test sets are provided in the repository for each resolution (with no overlap), so that additional DL models can be evaluated on the same data and compared with the results achieved in this study.</p> <p>The structure of the information shared in this repository is as follows:<br>The scripts have been grouped by tile resolution (256, 512 and 1024). First, the test set and the evaluation script can be found. For each tile resolution, there are two subfolders (corresponding to the "no overlap" and "12.5% overlap"). In each case, the Python scripts for training the models in the three repetitions are shared, and the trained models (H5 format) are shared in compressed form. Finally, for each resolution we also share the testing dataset which consists of two folders.</p> <p>The material is distributed under a CC-BY 4.0 license.</p>
Classification of blood cells dynamics with convolutional and recurrent neural networks: a sickle cell disease case study
<p>The fraction of red blood cells (RBC) adopting a specific motion under low shear flow is a promising inexpensive marker for monitoring the clinical status of patients with sickle cell disease (SCD). Its high-throughput measurement relies on the video analysis of thousands of cell motions for each blood sample to eliminate a large majority of unreliable samples(out of focus or overlapping cells) and discriminate between tank-treading and flipping motion, characterizing highly and poorly deformable cells respectively. These videos are of different durations (from 6 to more than 100 frames).</p> <p>This dataset contains four adult patients with SCD. They were enrolled in the study Drepaforme (approved by the institutional review board CPP Ouest 6 under the reference n°2018A00679-46) and were sampled weekly for several months. The movies were processed using in-house routines in Matlab (Matlab, R2016a) and RBC were detected individually and tracked over time. The database provided in this repository are already pre-processed sequences of tracked and centered RBC over time, each time step image being normalized to 31x31 pixels. Within the 32 experiments, the total number of sequences (or samples) is nearly 150 000. All sequences were semi-automatically labelled into 3 classes, depending on the dynamic of the cell: tank-treading, flipping and unreliable (140 000 are unreliable). The percentage of tank-treading cells with respect to all reliable cells (tank-treading+flipping) in every experiment is the final goal of this study.</p> <p>This dataset is very interesting to the community as it is a large database for cell dynamics classification: the class depends on the movement of the cell.</p> <p>An automatic processing of the database using a 2-stage deep learning model is available here https://github.com/icannos/redbloodcells_disease_classification</p> <p>For opening the data in python:</p> <p> from scipy.io import loadmat<br> x=loadmat('BG20191003shear10s01_Export.mat')</p> <p> * x['Norm_Tab'] is of size nb_samples x max_len_sequences x 31 x 31, where max_len_sequences is the length of the longest sequence of the series, typically ~150 to 180. The other sequences are padded with 31x31 zero matrices at the end in order to fill this maximal length.</p> <p> * x['Labels_Num'] is the corresponding label of each sequence, of size nb_samples. Label can be:<br> - 0 : "tank-treading" (or healthy)<br> - 1 : "flipping" (or tumbling, i.e. related to a SCD)<br> - 2 : "unreliable"</p>
Supplementary material for the paper "Detection and Characterization of Multiple Discontinuities in Cables with Time Domain Reflectometry and Convolutional Neural Networks"
<p>This archive contains supplementary material for the paper "Detection and Characterization of Multiple Discontinuities in Cables with Time Domain Reflectometry and Convolutional Neural Networks".</p> <p>In particular, the following data is provided:</p> <ul> <li>dataset of labeled simulated TDR signals (1 million samples)</li> <li>dataset of labeled experimental signals (16 samples)</li> <li>trained convolutional neural network model</li> </ul>
Using convolutional neural networks to efficiently extract immense phenological data from community science images
<p>Community science image libraries offer a massive, but largely untapped, source of observational data for phenological research. The iNaturalist platform offers a particularly rich archive, containing more than 49 million verifiable, georeferenced, open access images, encompassing seven continents and over 278,000 species. A critical limitation preventing scientists from taking full advantage of this rich data source is labor. Each image must be manually inspected and categorized by phenophase, which is both time-intensive and costly. Consequently, researchers may only be able to use a subset of the total number of images available in the database. While iNaturalist has the potential to yield enough data for high-resolution and spatially extensive studies, it requires more efficient tools for phenological data extraction. A promising solution is automation of the image annotation process using deep learning. Recent innovations in deep learning have made these open-source tools accessible to a general research audience. However, it is unknown whether deep learning tools can accurately and efficiently annotate phenophases in community science images. Here, we train a convolutional neural network (CNN) to annotate images of Alliaria petiolata into distinct phenophases from iNaturalist and compare the performance of the model with non-expert human annotators. We demonstrate that researchers can successfully employ deep learning techniques to extract phenological information from community science images. A CNN classified two-stage phenology (flowering and non-flowering) with 95.9% accuracy and classified four-stage phenology (vegetative, budding, flowering, and fruiting) with 86.4% accuracy. The overall accuracy of the CNN did not differ from humans (p = 0.383), although performance varied across phenophases. We found that a primary challenge of using deep learning for image annotation was not related to the model itself, but instead in the quality of the community science images. Up to 4% of A. petiolata images in iNaturalist were taken from an improper distance, were physically manipulated, or were digitally altered, which limited both human and machine annotators in accurately classifying phenology. Thus, we provide a list of photography guidelines that could be included in community science platforms to inform community scientists in the best practices for creating images that facilitate phenological analysis.</p>
Automatic taxonomic identification based on the Fossil Image Dataset (>415,000 images) and deep convolutional neural networks
<p>This is a Fossil Image Dataset, which contains >415000 images. A total of 50 clades were labeled, with a final 90% accuracy. We used the web crawler to download fossil images from the Internet. We declare that all the collected images are used for academic purposes only. If anyone wants to use this dataset, please agree on the Terms of access for the Fossil Image Dataset (FID). We uploaded two datasets: FID (contains 0.415 million images) and reduced-FID (60 thousand images, 1200 for each clade). Requirements of necessary preinstalled Python libraries, algorithms for analysis, and the model weights are available at <a href="https://github.com/XiaokangLiuCUG/Fossil_Image_Dataset">https://github.com/XiaokangLiuCUG/Fossil_Image_Dataset</a>.</p>
Convolutional Neural Networks for LPV-Approximations of Semi-discrete Navier-Stokes Equations
<pre><code>A `python` module with * a dynamic setup of *Convolutional Neural Networks* in `PyTorch` * an interface to `FEniCS` to generate data from FEM simulations of flows and * a numerical realization of FEM norm in the training neural networks developed to design very low-dimensional LPV approximations of incompressible Navier-Stokes equations.</code></pre> <p> </p> <p><code>These files contain the core module </code>and the scripts that produce the numerical examples of the paper with <a href="https://doi.org/10.3389/fams.2022.879140">doi:10.3389/fams.2022.879140</a></p> <p> </p> <pre><code>> Benner, Heiland, Bahmani (2022): *Convolutional Neural Networks for Very Low-dimensional LPV Approximations of Incompressible Navier-Stokes Equations* </code></pre> <p> </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.