BAMU · Final Year B.E. Artificial Intelligence and Machine Learning · CBCGS
Deep Learning
AIM403 · Semester VII · 4 credits · CT 20 + theory 80
Unit 1: History of Deep Learning and Perceptrons
10 hoursHistory and success stories of deep learning; the McCulloch–Pitts neuron and thresholding logic; perceptrons and their learning algorithm; multilayer perceptrons and their representation power; sigmoid neurons and gradient descent.
Examined in the 80-mark end-semester paper
- McCulloch–Pitts neuron and thresholding logic
- Perceptron learning rule
- Multilayer perceptron
- Gradient descent variants
Unit 2: Feed-Forward Networks and Optimisers
10 hoursFeed-forward networks and their representation power; backpropagation; momentum-based and Nesterov accelerated gradient descent, stochastic gradient descent, AdaGrad, RMSProp and Adam.
Examined in the 80-mark end-semester paper
- Feed-forward network architectures
- Backpropagation
- Momentum and Nesterov accelerated gradient descent
- Optimisers: AdaGrad, RMSProp and Adam
Unit 3: PCA, SVD and Autoencoders
10 hoursEigenvalues and eigenvectors, eigen-decomposition and bases; principal component analysis and its interpretations; singular value decomposition; autoencoders, their relation to PCA and their regularisation.
Examined in the 80-mark end-semester paper
- Eigen-decomposition, PCA and SVD
- Autoencoders and VAEs
- Autoencoders and their relation to PCA
Unit 4: Autoencoder Variants and Regularisation
10 hoursDenoising, sparse and contractive autoencoders; the bias–variance trade-off, L2 regularisation, early stopping, dataset augmentation, parameter sharing, noise injection, ensembles and dropout.
Examined in the 80-mark end-semester paper
- Denoising, sparse and contractive autoencoders
- Regularisation and dropout
- Early stopping, parameter sharing and ensembles
- Data augmentation
Unit 5: Word Representations and Convolutional Networks
10 hoursLearning vector representations of words; convolutional neural networks and LeNet, AlexNet, ZF-Net, VGGNet, GoogLeNet and ResNet; visualising networks with guided backpropagation, Deep Dream and Deep Art; fooling CNNs.
Examined in the 80-mark end-semester paper
- Word embeddings: Word2Vec and GloVe
- Convolution and feature maps
- CNN architectures: LeNet to ResNet
- Visualising CNNs: guided backpropagation and Deep Dream
Unit 6: Recurrent Networks and Attention
10 hoursRecurrent neural networks, backpropagation through time and truncated BPTT; vanishing and exploding gradients; GRU and LSTM; encoder–decoder models and the attention mechanism, including attention over images.
Examined in the 80-mark end-semester paper
- Recurrent neural networks
- Backpropagation through time
- Vanishing and exploding gradients
- LSTM
- GRU
- Encoder-decoder sequence models
- Attention mechanism
Unit contents are summarised from the official BAMU syllabus . AiCanCode.org is not affiliated with Dr. Babasaheb Ambedkar Marathwada University (BAMU).