AI-Powered Learning Platform

Subject Syllabus

AL503 (B) - Deep Learning

AIML · Semester 5

Course Modules

Learn topic-wise with Hyper AI and revise using mapped PYQs.

History of deep learning McCulloch-Pitts neuron Multilayer perceptrons (MLPs)Representation power of MLPs Sigmoid neurons Feedforward neural networks Backpropagation Weight initialization methods Batch normalization Representation learning GPU implementation Decomposition: PCA and SVD

Gradient descent (GD)Momentum-based gradient descent Nesterov accelerated gradient descent Stochastic gradient descent AdaGrad, Adam, and RMSProp Autoencoders Regularization in autoencoders Denoising autoencoders Sparse autoencoders Contractive autoencoders Variational autoencoder Relationship of autoencoders with PCA and SVD Dataset augmentation

Introduction to convolutional neural networks and their architectures CNN terminologies: ReLU activation function, stride, padding, and pooling Convolution operations and convolutional kernels Types of layers: convolutional, pooling, and fully connected Visualizing CNNs CNN examples: LeNet, AlexNet, ZF-Net, VGGNet, GoogLeNet, ResNet, and RCNN Deep Dream and Deep Art Regularization: dropout, DropConnect, unit pruning, and stochastic pooling Artificial data, injecting noise in the input, and early stopping Limiting the number of parameters and weight decay

Introduction to deep recurrent neural networks and their architectures Backpropagation through time (BPTT)Vanishing and exploding gradients Truncated BPTT Gated recurrent units (GRUs)Long short-term memory (LSTM)Solving the vanishing gradient problem with LSTMs Encoding and decoding in an RNN Attention mechanism Attention over images Hierarchical attention Directed graphical models Applications of deep RNNs in image processing, natural language processing, speech recognition, and video analytics

Introduction to deep generative models Restricted Boltzmann machines (RBMs)Gibbs sampling for training RBMs Deep belief networks Markov networks and Markov chains Autoregressive models: NADE, MADE, and PixelRNN Generative adversarial networks (GANs)Applications of deep learning in object detection, speech and image recognition, video analysis, NLP, and medical science