Find research datasets worth reusing

opencc-by-4.0Dec 2021View details →

A Probabilistic Framework for Mutation Testing in Deep Neural Networks - Models archive Part 2

Models used as part of the paper "A Probabilistic Framework for Mutation Testing in Deep Neural Networks ?" submitted to the journal Information and Software Technology Replication package using the data is available at https://github.com/FlowSs/PM

opencc-by-4.0Dec 2021View details →

A Probabilistic Framework for Mutation Testing in Deep Neural Networks - Models archive Part 3

opencc-by-4.0Dec 2021View details →

A Probabilistic Framework for Mutation Testing in Deep Neural Networks - Models archive Part 1

opencc-by-4.0Jan 2022View details →

Predicting dry matter intake in Canadian Holstein dairy cattle using milk mid-infrared reflectance spectroscopy and other commonly available predictors via artificial neural networks

Supplementary Tables

Model Zoo: A Dataset of Diverse Populations of Neural Network Models - STL10 - Preprocessed Datasets

Abstract In the last years, neural networks have evolved from laboratory environments to the state-of-the-art for many real-world problems. Our hypothesis is that neural network models (i.e., their weights and biases) evolve on unique, smooth trajectories in weight space during training. Following, a population of such neural network models (refereed to as “model zoo”) would form topological structures in weight space. We think that the geometry, curvature and smoothness of these structures contain information about the state of training and can be reveal latent properties of individual models. With such zoos, one could investigate novel approaches for (i) model analysis, (ii) discover unknown learning dynamics, (iii) learn rich representations of such populations, or (iv) exploit the model zoos for generative modelling of neural network weights and biases. Unfortunately, the lack of standardized model zoos and available benchmarks significantly increases the friction for further research about populations of neural networks. With this work, we publish a novel dataset of model zoos containing systematically generated and diverse populations of neural network models for further research. In total the proposed model zoo dataset is based on six image datasets, consist of 24 model zoos with varying hyperparameter combinations are generated and includes 47’360 unique neural network models resulting in over 2’415’360 collected model states. Additionally, to the model zoo data we provide an in-depth analysis of the zoos and provide benchmarks for multiple downstream tasks as mentioned before. Dataset This dataset is part of a larger collection of model zoos and contains the zoos trained on the labelled samples from STL10. All zoos with extensive information and code can be found at www.modelzoos.cc. This repository contains the preprocessed model zoos wrapped in a custom pytorch dataset class (filenames beginning with "dataset"). Zoos are trained with small and large CNN models, in three configurations varying the seed only (seed), varying hyperparameters with fixed seeds (hyp_fix) or varying hyperparameters with random seeds (hyp_rand). Due to the large filesize, the raw datasets are hosted in a separate repository. The index_dict.json files contain information on how to read the vectorized models. For more information on the zoos and code to access and use the zoos, please see www.modelzoos.cc.

Model Zoo: A Dataset of Diverse Populations of Neural Network Models - USPS

Abstract In the last years, neural networks have evolved from laboratory environments to the state-of-the-art for many real-world problems. Our hypothesis is that neural network models (i.e., their weights and biases) evolve on unique, smooth trajectories in weight space during training. Following, a population of such neural network models (refereed to as “model zoo”) would form topological structures in weight space. We think that the geometry, curvature and smoothness of these structures contain information about the state of training and can be reveal latent properties of individual models. With such zoos, one could investigate novel approaches for (i) model analysis, (ii) discover unknown learning dynamics, (iii) learn rich representations of such populations, or (iv) exploit the model zoos for generative modelling of neural network weights and biases. Unfortunately, the lack of standardized model zoos and available benchmarks significantly increases the friction for further research about populations of neural networks. With this work, we publish a novel dataset of model zoos containing systematically generated and diverse populations of neural network models for further research. In total the proposed model zoo dataset is based on six image datasets, consist of 24 model zoos with varying hyperparameter combinations are generated and includes 47’360 unique neural network models resulting in over 2’415’360 collected model states. Additionally, to the model zoo data we provide an in-depth analysis of the zoos and provide benchmarks for multiple downstream tasks as mentioned before. Dataset This dataset is part of a larger collection of model zoos and contains the zoos trained on the labelled samples from USPS. All zoos with extensive information and code can be found at www.modelzoos.cc. This repository contains two types of files: the raw model zoos as collections of models (file names beginning with "usps_"), as well as preprocessed model zoos wrapped in a custom pytorch dataset class (filenames beginning with "dataset"). Zoos are trained in three configurations varying the seed only (seed), varying hyperparameters with fixed seeds (hyp_fix) or varying hyperparameters with random seeds (hyp_rand). The index_dict.json files contain information on how to read the vectorized models. For more information on the zoos and code to access and use the zoos, please see www.modelzoos.cc.

Dataset: Gauze detection and segmentation in minimally invasive surgery video using convolutional neural networks

Dataset of the Gauze detection and segmentation in minimally invasive surgery video using convolutional neural networks article. Further information is available in the README file.

opencc-by-4.0Nov 2021View details →

PINNup: Robust neural network wavefield solutions using frequency upscaling and neuron splitting

Solving for the frequency-domain scattered wavefield via physics-informed neural network (PINN) has great potential in increasing the flexibility and reducing the computational cost of seismic modeling and inversion. We propose a novel implementation of PINN using frequency upscaling and neuron splitting, which allows the neural network model to grow in size as we increase the frequency while leveraging the information from the pre-trained model for lower-frequency wavefields, resulting in fast convergence to high-accuracy wavefield solutions. In this letter, we present the relevant dataset to the paper.

Ice water path retrievals from Meteosat-9 with quantile regression neural networks: video supplement

Supplementary videos used from in A. Amell, P. Eriksson, S. Pfreundschuh: Ice water path retrievals form Meteosat-9 with quantile regression neural networks.

opencc-by-sa-4.0Apr 2022View details →

Impact of training set size on the ability of deep neural networks to deal with label noise

This is the data set accompanying the publication 'Impact of training set size on the ability of deep neural networks to deal with label noise'   License This work is shared under the <code>Creative Commons Attribution ShareAlike 4.0 International License</code><code> (CC BY-SA 4.0)</code>: https://creativecommons.org/licenses/by-sa/4.0/ The dataset is taken from the 6th SpaceNet challenge (https://spacenet.ai/sn6-challenge/) which also falls under the <code>CC BY-SA 4.0 license</code>. The original dataset was modified by removing objects from the buildings json files and converting the json files to png images. Creators of the SpaceNet dataset: Shermeyer, J., Hogan, D., Brown, J., Etten, A.V., Weir, N., Pacifici, F., Hänsch, R., Bastidas, A., Soenen, S., Bacastow, T.M., & Lewis, R. SpaceNet partners: https://spacenet.ai/about-us/

opencc-by-4.0Jul 2022View details →

Neural network training sets

This dataset contains the synthetically generated particle image datasets used to train a denoising autoencoder (DAE_synthetic_dataset.zip) and a fully convolutional network for semantic segmentation (semanticSegmentation_synthetic_dataset.zip). The necessary models and maps used to generate the dataset are included in synthetic_volumes_and_pdbs.zip. Lastly, the trained neural networks are included in trained_neural_networks.zip.

openNov 2021View details →

Accelerating DNA-PAINT imaging with a deep neural network

(<a href="https://zenodo.org/api/files/7030c526-533f-44ce-8b8c-01a979fd6b6f/RawFrames_P5-IS_20pM.tif?versionId=55a12821-8472-42f5-91b5-eabda1516e2b">RawFrames_P5-IS_20pM.tif</a>) Raw DNA-PAINT SMLM frames with isolated emitters taken on TOM20 labelled MNTB neuronal rat tissue with P5 imaging strand at 20 pM concentration.  (<a href="https://zenodo.org/api/files/7030c526-533f-44ce-8b8c-01a979fd6b6f/Binned_HighDensity_30kpatches.tif?versionId=ac01dd00-3c44-4059-a1b1-3b1045fa636d">Binned_HighDensity_30kpatches.tif</a> and <a href="https://zenodo.org/api/files/7030c526-533f-44ce-8b8c-01a979fd6b6f/Binned_HighDensity_30kpatches.csv?versionId=31e792fc-fab5-43cd-af81-a271595586d0">Binned_HighDensity_30kpatches.csv</a>) Artificially summed high-density emitter patches with the corresponding emitter coordinates used for training the DeepSTORM neural network.   (DeepSTORM_model_metadata.mat and DeepSTORM_model_weights_best.hdf5) The trained model metadata and weights used for all predicted images in the study. (Figure 4_Large_super-resolution_image.png) The large super-resolution image in Figure 4 and Figure S4. (High-density-frames_Images 1 - 5 .tif) Five high-emitter density raw frames for alpha-tubulin and TOM20 of 400 frames each. (<a href="https://zenodo.org/api/files/b98093d2-f689-4fef-8222-220c07ca4fe9/Ground-truth-rendered_Image4_Tubulin.tif">Ground-truth-rendered_Image 1 - 5</a>.tif) Five ground truth images rendered in Picasso (drift-corrected, linked localisations, pixel size 13.37 nm/pixel) for alpha-tubulin and TOM20. (<a href="https://zenodo.org/api/files/e726ff9b-45ba-4d3a-b3ba-3e6eaca03d35/Low%20density%20frames%20for%20GT%20images.zip">Low density frames for GT images.zip</a>) Five low-emitter density (0.5 nM) raw frames for alpha-tubulin and TOM20 of 10000 frames each which were used to render ground truth images. (<a href="https://zenodo.org/api/files/e726ff9b-45ba-4d3a-b3ba-3e6eaca03d35/Bassoon_Homer_datasets.zip">Bassoon_Homer_datasets.zip</a>) Three Bassoon and Homer datasets each with a ground truth image and the corresponding high-density raw frames (5 nM, 800 frames).

opencc-zeroAug 2022View details →

Supplementary material 1 from: Behei N, Tryhubchak O, Pryymak B (2022) Development of amlodipine and enalapril combined tablets based on quality by design and artificial neural network for confirming of qualitative composition. Pharmacia 69(3): 779-789. https://doi.org/10.3897/pharmacia.69.e86876

The results of the study of pharmaco-technological parameters of intermediates and amlodipine tablets with enalapril, data of the functions of desirability

opencc-by-4.0Sep 2022View details →

Machine learning for Material Science 2022 - Neural Networks Assignment Dataset

Data for the neural network assignment

Fig. 5 in Pretrained Convolutional Neural Networks Perform Well in a Challenging Test Case: Identification of Plant Bugs (Hemiptera: Miridae) Using a Small Number of Training Images

Fig. 5. Visualization of a feature with high importance and a feature with low importance from a configuration B. The importance of these features for the identification accuracy was determined using permutation tests (see the methods). For each taxon or group (rows) several randomly selected specimens (columns) are shown. For the two selected Global Average Pooling layer features, the corresponding features of the preceding (Max Pooling) layer are visualized as those show specific image parts that had higher activations.Yellow represents the maximal activation strength; dark blue represents the minimal activation strength.

Fig. 2 in Pretrained Convolutional Neural Networks Perform Well in a Challenging Test Case: Identification of Plant Bugs (Hemiptera: Miridae) Using a Small Number of Training Images

Fig. 2. Schematic representation of the networks used. On top are the convolutional layers of VGG16, grouped into five blocks. Output of each Max Pooling layer is fed into Global Average Pooling layer. Numbers near each block name indicate number of features in the Global Average Pooling layer. Height of layers roughly corresponds to resolution (except for Global Average Pooling layer), while width roughly corresponds to the number of feature maps or features produced.Then, in approach A, outputs of five blocks are concatenated and passed to the linear classifier. In approach B, output of only one block (block 3 in the final configuration) is passed to the linear classifier. In approach C CNN outputs are as in approach A, but instead connected to a DNN with two layers of 320 fully connected (FC) neurons followed by a prediction layer (PL), with number of neurons equal to number of species classified. Finally, approach D features CNN as in approach B which is connected to DNN as in approach C.

Fig. 1 in Pretrained Convolutional Neural Networks Perform Well in a Challenging Test Case: Identification of Plant Bugs (Hemiptera: Miridae) Using a Small Number of Training Images

Fig. 1. Dorsal habitus photos of males and females of Tuxedo spp., Pygovepres vaccinicola, and Phallospinophylus setosus generated for and used in this study.

Fig. 4 in Pretrained Convolutional Neural Networks Perform Well in a Challenging Test Case: Identification of Plant Bugs (Hemiptera: Miridae) Using a Small Number of Training Images

Fig. 4. Validation accuracy and accuracy on test data for SVM linear classifier (A, B) and DNN approaches (C, D) for the three datasets.

Fig. 3 in Pretrained Convolutional Neural Networks Perform Well in a Challenging Test Case: Identification of Plant Bugs (Hemiptera: Miridae) Using a Small Number of Training Images

Fig. 3. Identification accuracy for the male (top) and female (bottom) Tuxedo dataset for the five selected resolutions and the individual blocks 1–5 and the concatenated block.