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 networks”
A "short blanket" dilemma for a state-of-the-art neural network potential for water: Reproducing experimental properties or the underlying many-body physics?
<p>Deep neural network (DNN) potentials have recently gained popularity in computer simulations of a wide range of molecular systems, from liquids to materials.<br> In this study, we explore the possibility of combining the computational efficiency of the DeePMD framework and the demonstrated accuracy of the MB-pol data-driven many-body potential to train a DNN potential for large-scale simulations of water across its phase diagram.<br> We find that the DNN potential is able to reliably reproduce the MB-pol results for liquid water but provides a less accurate description of the vapor-liquid equilibrium properties.<br> This shortcoming is traced back to the inability of the DNN potential to correctly represent many-body interactions.<br> An attempt to explicitly include information about many-body effects results in a new DNN potential that exhibits the opposite performance, being able to correctly reproduce the MB-pol vapor-liquid equilibrium properties but losing accuracy in the description of the liquid properties.<br> These results suggest that DeePMD-based DNN potentials are not able to correctly "learn" and, consequently, represent many-body interactions, which implies that DNN potentials may have limited ability to predict properties for state points that are not explicitly included in the training process.<br> The computational efficiency of the DeePMD framework can still be exploited to train DNN potentials on data-driven many-body potentials, which can thus enable large-scale, "chemically accurate" simulations of various molecular systems, with the caveat that the target state points must have been adequately sampled by the reference data-driven many-body potential in order to guarantee a faithful representation of the associated properties.</p>
Prediction of inter-chain distance maps of protein complexes with 2D attention-based deep neural networks
<p>Residue-residue distance information is useful for predicting tertiary structures of protein monomers or quaternary structures of protein complexes. Many deep learning methods have been developed to predict intra-chain residue-residue distances of monomers accurately, but few methods can accurately predict inter-chain residue-residue distances of complexes. We develop a deep learning method CDPred (i.e., Complex Distance Prediction) based on the 2D attention-powered residual network to address the gap. Tested on two homodimer datasets, CDPred achieves the precision of 60.94% and 42.93% for top L/5 inter-chain contact predictions (L: length of the monomer in homodimer), respectively, substantially higher than DeepHomo’s 37.40% and 23.08% and GLINTER’s 48.09% and 36.74%. Tested on the two heterodimer datasets, the top Ls/5 inter-chain contact prediction precision (Ls: length of the shorter monomer in heterodimer) of CDPred is 47.59% and 22.87% respectively, surpassing GLINTER’s 23.24% and 13.49%. Moreover, the prediction of CDPred is complementary with that of AlphaFold2-multimer.</p>
ArrayCGH microarray images for 'Autoencoder and NCA based neural network model to estimate survival prognosis in multiple myeloma using arrayCGH data'
<p>ArrayCGH microarray images for 'Autoencoder and NCA based neural network model to estimate survival prognosis in multiple myeloma using arrayCGH data'</p>
Core-loss EELS dataset and neural networks for element identification
<p>We present a large dataset containing simulated core-loss electron energy loss spectroscopy (EELS) spectra with the elemental content as ground-truth labels. Additionally we present some neural networks trained on this data for element identification. </p> <p>The simulated dataset contains zero padded core-loss spectra from 0 to 3072 eV, which represents 107 core-loss edges through all 80 elements from Be up to Bi. The core-loss edges are calculated from the generalised oscillator strength (GOS) database presented by Zhang et al.[1] Generic fine structures using lifetime broadened peaks are used to imitate fine structure due to solid-state effects in experimental spectra. Generic low-loss regions are used to imitate the effect of multiple scattering. Each spectrum contains at least one edge of a given query element and possibly additional edges depending on samples drawn from The Materials Project [2]. The dataset contains for each of the 80 elements: 7000 training spectra, 1500 test spectra, 600 validation spectra and 100 spectra representing only the query element. This results in a total 736 000 labeled spectra.</p> <p>Code on how to <br> - read the simulated data<br> - transform HDF5 format to TFRecord format<br> - train and evaluate neural networks using the simulated data<br> - use the trained networks for automated element identification<br> is available on GitHub at arnoannys/EELS_ID</p> <p>A full report on the simulation of the dataset and the training and evaluation of the neural networks can be found at: Annys, A., Jannis, D. & Verbeeck, J. Deep learning for automated materials characterisation in core-loss electron energy loss spectroscopy. <em>Sci Rep</em> 13, 13724 (2023). https://doi.org/10.1038/s41598-023-40943-7</p> <p>[1] Zezhong Zhang, Ivan Lobato, Daen Jannis, Johan Verbeeck, Sandra Van Aert, & Peter Nellist. (2023). Generalised oscillator strength for core-shell electron excitation by fast electrons based on Dirac solutions (1.0) [Data set]. Zenodo. https://doi.org/10.5281/zenodo.7729585<br> [2] Anubhav Jain, Shyue Ping Ong, Geoffroy Hautier, Wei Chen, William Davidson Richards, Stephen Dacek, Shreyas Cholia, Dan Gunter, David Skinner, Gerbrand Ceder, Kristin A. Persson; Commentary: The Materials Project: A materials genome approach to accelerating materials innovation. __APL Mater__ 1 July 2013; 1 (1): 011002. [https://doi.org/10.1063/1.4812323](https://doi.org/10.1063/1.4812323)</p>
Current Harmonics Minimization of PMSM Based on Iterative Learning Control and Neural Networks: Motor Data
<p>The provided motor data corresponds to an electrical machine with 24 stator slots and 16 poles. As is common in electrical machines, this motor generates unwanted flux and current harmonics. However, the accompanying paper presents an effective solution to suppress these harmonics through the combined use of Iterative Learning Control (ILC) and Neural Networks (NNs).</p> <p>The ILC method demonstrates proficient compensation for harmonics during operations with constant speed and current reference values. Additionally, Neural Networks are trained with data derived from ILC, proving to be highly effective in suppressing harmonics even during transient operation. The simulation model used in the study is based on flux and torque maps, dependent on dq-currents and the electrical angle. These maps are obtained from Finite Element Method (FEM) simulations of an interior permanent magnet synchronous machine (IPM) and are openly published here, intended to facilitate other researchers in making direct comparisons with their own methodologies.</p> <p>Simulation results presented in the paper confirm that the integration of ILC and NNs leads to superior elimination of current harmonics during transient operations compared to using ILC alone.<br> If you use the provided maps and motor data, kindly cite the associated paper for reference: https://doi.org/10.3390/machines11080784, https://www.mdpi.com/2075-1702/11/8/784</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>
Datasets for Paper "BenchTemp: A General Benchmark for Evaluating Temporal Graph Neural Networks"
<p>Datasets for Paper "BenchTemp: A General Benchmark for Evaluating Temporal Graph Neural Networks"<br> URL: https://github.com/qianghuangwhu/benchtemp</p> <p>Openreview: https://openreview.net/forum?id=rnZm2vQq31</p> <p><br> There are 19 (15+4) benchmark temporal graph datasets:<br> reddit,<br> wikipedia,<br> mooc,<br> lastfm,<br> enron,<br> SocialEvo,<br> uci,<br> CollegeMsg,<br> TaobaoSmall,<br> CanParl,<br> Contacts,<br> Flights,<br> UNtrade,<br> USLegis,<br> UNvote,</p> <p>DGraphFin,</p> <p>TaobaoLarge,</p> <p>YoutubeReddit,</p> <p>YoutubeRedditLarge</p> <p> </p> <p><br> Each dataset has three files:<br> 1. ml_{data_name}.csv - the csv file of the Temporal Graph.</p> <p>This file have five columns with properties:</p> <p>'u': the id of the user.<br> 'i': the id of the item.<br> 'ts': the timestamp of the interaction (edge) between the user and the item.<br> 'label': the label of the interaction (edge).<br> 'idx': the index of the interaction (edge).<br> For example:</p> <p>,u,i,ts,label,idx<br> 0,1,2,0.0,0.0,1<br> 1,1,3,0.0,0.0,2<br> 2,1,4,0.0,0.0,3<br> 2. ml_{data_name}.npy - the edge features corresponding to the interactions (edges) in the the Temporal Graph..</p> <p>3. ml_{data_name}_node.npy - the initialization node features of the Temporal Graph.</p>
BLM-AgrF: A New French Benchmark to Investigate Generalization of Agreement in Neural Networks
<p>BLM-AgrF is a French dataset for learning the underlying rules of subject-verb agreement in sentences, developed in the BLM framework, a new task inspired by visual IQ tests known as Raven's Progressive Matrices. In this task, an instance consists of sequences of sentences with specific attributes. To predict the correct answer as the next element of the sequence, a model must correctly detect the generative model used to produce the dataset.</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>
Glassware images and code samples for training and identification of glassware by neural networks
<p>These images were used to perform an image identification exercise with first-year students in the author's Introduction to Scientific Computing course. The zip file also includes sample Mathematica notebooks that were used to perform the training and data analysis of the neural network's performance. The corresponding publication in the Journal of Computational Science Education can be found here <a href="https://doi.org/10.22369/issn.2153-4136/12/1/2">https://doi.org/10.22369/issn.2153-4136/12/1/2</a></p>
DeepBedMap: A super-resolution neural network created bed topography of Antarctica
<p>Going beyond BEDMAP2 using a super resolution deep neural network.</p> <p>deepbedmap_v1.1.0.zip: Python code for the DeepBedMap Super-Resolution Generative Adversarial Network.</p> <p>deepbedmap_dem.tif: Digital Elevation Model (250 m spatial resolution) in GeoTiff format, using Antarctic Polar Stereographic Projection (EPSG:3031).</p> <p>srgan_generator_model_weights.npz: The Generator neural network weights/parameters as a NumPy zip file.</p> <p> </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>
Large-Scale Gravitational Lens Modeling with Bayesian Neural Networks for Accurate and Precise Inference of the Hubble Constant - Datasets, Trained Models, BNN Samples, and MCMC Chains
<p>We publish the training/validation/test datasets, trained model weights, configuration files, Bayesian neural network samples, and MCMC chains used to produce the figures in the LSST DESC paper, "Large-Scale Gravitational Lens Modeling with Bayesian Neural Networks for Accurate and Precise Inference of the Hubble Constant." They are formatted to be used with the DESC package "H0rton" (<a href="https://github.com/jiwoncpark/h0rton">https://github.com/jiwoncpark/h0rton</a>). Additional descriptions can be found in the README. Please contact Ji Won Park (@jiwoncpark) on GitHub or <a href="https://github.com/jiwoncpark/h0rton/issues">make an issue</a> for any questions.</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>
Semi-Recurrent Neural Networks In IllustrisTNG And N-Body Simulations
<p>This is the official data repository for the MNRAS publication <a href="https://arxiv.org/abs/2203.12702">Modelling the galaxy-halo connection using semi-recurrent neural networks</a>, and subsequent works <a href="https://arxiv.org/abs/2409.16548">Optimised neural network predictions of galaxy formation histories using semi-stochastic corrections</a> and <a href="https://arxiv.org/abs/2409.16079">Evaluating the galaxy formation histories predicted by a neural network in pure dark matter simulations</a>. For details on access and utilisation of the data and code, see documentation.pdf in the affiliated <a href="https://github.com/hgc4/TNG-Networks">GitHub repository</a>.</p>
Dataset for manuscript 'CeyeHao: AI-driven microfluidic flow programming with hierarchically assembled obstacles in microchannel and receptive-field-augmented neural network'
<p>This dataset contains:<br>1. The dataset used to train the models related to the manuscript 'CeyeHao: AI-driven microfluidic flow programming using hierarchically assembled obstacles in microchannel with receptive-field-augmented neural network'.<br>2. A checkpoint of trained 'CEyeNet' proposed in the manuscript.<br>3. Example microchannels designed in the manuscript to produce semantic flow profiles</p> <p>This dataset is intended for research and academic purpose.</p> <p>Detailed description please refer to the enclosed ReadMe.txt.</p>
Data package for paper "DeepGlow: an efficient neural-network emulator of physical afterglow models for gamma-ray bursts and gravitational-wave events
<p>This is a data package accompanying the paper "DeepGlow: an efficient neural-network emulator of physical afterglow models for gamma-ray bursts and gravitational-wave events".</p>
Dataset from the paper entitled "Complex structure of molten FLiBe (2 LiF – BeF2) examined by experimental neutron scattering, X-ray scattering, and deep neural network-based molecular dynamics"
<p>Dataset from the paper entitled "Complex structure of molten FLiBe (2 LiF – BeF2) examined by experimental neutron scattering, X-ray scattering, and deep neural network-based molecular dynamics". These data include experimental total scattering measurements and molecular dynamics simulations on the molten structure of FLiBe. </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>
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.