Deep Learning & Applications

This repo contains the codes, reports, datasets and question papers for the assignments in the course CS671 Deep Learning & Applications 2022 February-May Semester at Indian Institute of Technology, Mandi. Syllabus • Moodle

Perceptron & Fully Connected Deep Neural Network - DNN (Classification & Regression)

1. Implementation of Perceptron from scratch for classification tasks view
2. Implementation of Fully Connected Neural Network (FCNN) with one hidden layer from scratch for classification tasks view
3. Implementation of Fully Connected Neural Network (FCNN) with two hidden layer from scratch for classification tasks view
4. Implementation of Perceptron from scratch for regression tasks view
5. Implementation of Fully Connected Neural Network (FCNN) with one hidden layer from scratch for regression tasks view
6. Implementation of Fully Connected Neural Network (FCNN) with two hidden layer from scratch for regression tasks view

Perceptron & FCNN Report view • Question Paper open • Dataset download

Study of Various Optimizers {Tasks using Handwritten digits (MNIST digit dataset)}

1. FCNN with 3 hidden layers using cross entropy loss by stochastic gradient descent (SGD) algorithm view
2. FCNN with 3 hidden layers using cross entropy loss by vanilla gradient descent algorithm view
3. FCNN with 3 hidden layers using cross entropy loss by SGD with momentum (NAG) view
4. FCNN with 3 hidden layers using cross entropy loss by RMSProp algorithm view
5. FCNN with 3 hidden layers using cross entropy loss by Adam optimizer view

Autoencoders

1. One hidden layer autoencoder for compression view
2. Three hidden layer autoencoder for compression view
3. One hidden layer autoencoder for classification task view
4. Three hidden layer autoencoder for classification task view
5. Denoising Autoencoder view

Optimizers and Autoencoders Report view • Question Paper open • Dataset download

Convolutional Neural Network - CNN

1. 2 dimensional convolution (with zero padding & stride) from scratch view
2. Building Convolutional Layers view
3. Building Network and Backpropagation view
4. Using Builtin functions view
5. GradCAM - Visualization of the Influence of Input Pixels view
6. VGG19 Model view
7. Visualizing the patches that maximally activate a neuron (VGG16) view

Convolutional Neural Network Report view • Question Paper open • Dataset download

Recurrent Neural Network - RNN

1. Preprocessing of data sets view
2. Implementation of RNN using Keras and PyTorch (Both datasets) view

Long Short Term Memory - LSTM

1. Preprocessing of data sets view
2. Implementation of LSTM using PyTorch view

RNN and LSTM report view • Question Paper open • Dataset download

Tutorials [Jan-May 2023: TA]

1. Tutorial I: Introduction to keras, DNNs & Autoencoders in keras
2. Tutorial II: CNNs: Plotting Feature maps, kernels, Transfer learning, Visualisation of patches maximally activate a neuron see
3. Tutorial III: RNNs & LSTMs: Passing variable length inputs, more