Mumbai University · Final Year Electronics and Telecommunication Engineering · Rev-2019 'C' Scheme
Deep Learning
ECCDLO7012 · Department Optional Course 3 · Semester VII · 3 credits · Internal tests 20 (average of two) + end-semester 80
End-semester paper (3 hours): Six questions of 20 marks; Q1 is compulsory and you answer four. Each module's share of the paper follows its lecture hours.
One of the Department Optional Course 3 options.
Module 1: Introduction to Deep Learning
3 hoursHistory of deep learning; deep feedforward networks; gradient-based learning and hidden units; architecture design; backpropagation.
Examined in the 80-mark end-semester paper (about 8% of it, by lecture hours); the class tests cover what has been taught by then
- Deep feedforward networks
- Backpropagation algorithm
Module 2: TensorFlow for Deep Learning
6 hoursComputational graphs; regression and gradient descent in TensorFlow; TensorBoard; Keras; preprocessing and augmenting image data.
Examined in the 80-mark end-semester paper (about 15% of it, by lecture hours); the class tests cover what has been taught by then
- TensorFlow basics
- Keras and PyTorch
- Image preprocessing and augmentation
Module 3: Regularisation and Optimisation
6 hoursL1, L2, early stopping and dropout; challenges in optimisation; initialisation; AdaGrad, RMSProp and Adam; batch normalisation.
Examined in the 80-mark end-semester paper (about 15% of it, by lecture hours); the class tests cover what has been taught by then
- Overfitting, dropout and regularisation
- Optimisers: AdaGrad, RMSProp and Adam
- Batch normalisation
Module 4: Evolution of CNNs
8 hoursLeNet, AlexNet, VGG, GoogLeNet, ResNet and U-Net compared; evaluation; image classification and object detection.
Examined in the 80-mark end-semester paper (about 21% of it, by lecture hours); the class tests cover what has been taught by then
- CNN architectures: LeNet, AlexNet and VGG
- ResNet and DenseNet
- GoogLeNet and U-Net
- Image classification and detection
Module 5: Sequence Modelling
8 hoursRecurrent, recursive and bidirectional nets; encoder–decoder architectures; LSTM and GRU; image captioning and time-series forecasting.
Examined in the 80-mark end-semester paper (about 21% of it, by lecture hours); the class tests cover what has been taught by then
- Recurrent neural networks
- LSTM and GRU
- Sequence-to-sequence models
Module 6: Encoder–Decoder Models
8 hoursAutoencoders and manifold learning; sparse, denoising and contractive autoencoders; deep autoencoders; variational autoencoders and the reparameterisation trick; applications.
Examined in the 80-mark end-semester paper (about 21% of it, by lecture hours); the class tests cover what has been taught by then
- Autoencoders and RBMs
- Sparse, denoising and contractive autoencoders
- Variational autoencoders
Module contents are summarised from the official Mumbai University syllabus . AiCanCode.org is not affiliated with the University of Mumbai.