sudipto153 Goto Github PK

clustering

The code creates a network of given number of clusters from a given dataset.

cnn_pytorch_fashionmnist

# CNN_pytorch_FashionMNIST The CNN is built using CNN. It uses the FashionMNIST dataset which was built on the platform of MNIST dataset. The dataset contains a collection of 70,000 pictures (1, 28, 28) of dress, foot-wears etc. The CNN achieves nearly 97% accuracy in training set and more than 90% accuracy in test set with 30 epochs trained on one GPU. The CNN uses three convolution layers and three fully connected layers. Two special classes - RunBuilder & RunManager were built to test various netoworks with different set of hyperparameters. These classes have also been incorporated in the code. Any new parameter can be easily added using the params dictionary and making few changes. The RunManager class stores the values for each run which can be used in tensorboard to visualize the impact of changing hyperparameters. The results are also saved in a .csv and .json file to be utilized later. This class also enables the user to visualize the accuracy with respect to parameters in real time while training.

covid_detecting_cnn

The model can detect the x-ray images of the lungs of COVID-19 patients with reasonable accuracy. The validation accuracy and test accuracy after training for ten epochs are 95.63% and 96.44% respectively. The model consists of a Residual Network Architecture with 11 layers.

data_visualization

dress_gan

The Generative Adversarial Network consists of a generator and a discriminator which are made up of small residual network architecture. It started giving out quite good results after training for only 20 epochs.

human-protein-multilabel-classification

The model was built on top the 'Resnext50_32x4d' using transfer learning. It can detect ten types of human protein structures in a given picture.

omicsgat

Graph Attention Network for Cancer Subtype Analyses

pytorchgeometrictutorial

Pytorch Geometric Tutorials

resnet_pytorch_cifar10

The simple Resnet (Residual Neural Architecture) is a model consisting of 15 layers. It was trained on the CIFAR10 dataset which is regarded as the basic testbench for CNNs.

stgat

Integrating Spatial Transcriptomics and Bulk RNA-seq: Predicting Gene Expression with Enhanced Resolution through Graph Attention Networks

tbt_asr_enhanced

This is an enhanced ASR version of TBT attack method on DNNs. It can classify the given input images to a target class with more than 90% success rate with flipping only 10 bits.

transformer-model-implementation

This is an implementation of the transformer model from 'Attention is All You Need' paper as a part of a course project.