SPPU · Final Year Computer Engineering · 2019 Course (Revised)
Deep Learning
410251 · Semester VIII · 3 credits · In-semester 30 + end-semester 70
Unit 1: Foundations of Deep Learning
7 hoursML and DL, supervised and unsupervised learning, bias–variance, hyperparameters, regularisation; limits of ML; history, advantages and challenges of DL; representation learning, architecture design principles, applications; TensorFlow, Keras, PyTorch, Caffe and Shogun.
Marks: 15 in the in-semester exam
- Deep learning vs machine learning
- History, advantages and challenges of deep learning
- Representation learning and deep architectures
- Deep learning frameworks: TensorFlow, Keras and PyTorch
Unit 2: Deep Neural Networks (DNNs)
7 hoursThe biological neuron, perceptron, multilayer feed-forward networks; forward and backpropagation; activation functions; loss functions for regression, classification and reconstruction; learning rate, regularisation, momentum, sparsity; XOR; gradient-based learning; vanishing and exploding gradients; sentiment analysis with PyTorch.
Marks: 15 in the in-semester exam
- Perceptron and feed-forward networks
- Forward and backpropagation
- Activation functions in deep networks
- Loss functions
- Hyperparameters: learning rate, momentum and regularisation
- Gradient descent and vanishing and exploding gradients
Unit 3: Convolution Neural Network (CNN)
7 hoursArchitecture overview; padding, strides, ReLU, pooling, fully connected layers, interleaving layers, local response normalisation; training a CNN.
Marks: 17–18 in the end-semester exam
- CNN architecture
- Convolution, padding and strides
- Pooling and fully connected layers
- Training a CNN
Unit 4: Recurrent and Recursive Nets
7 hoursUnfolding computational graphs, RNNs, bidirectional RNNs, encoder–decoder sequence-to-sequence models, deep and recursive networks, long-term dependencies, echo state networks, leaky units, LSTM and gated RNNs; practical methodology: metrics, baselines, more data, hyperparameters.
Marks: 17–18 in the end-semester exam
- Recurrent neural networks
- Bidirectional and deep RNNs
- Encoder–decoder sequence-to-sequence models
- Long-term dependencies, LSTM and GRU
- Recursive networks and echo state networks
- Practical methodology: metrics, baselines and hyperparameters
Unit 5: Deep Generative Models
8 hoursBoltzmann machines, deep belief networks; GANs, the discriminator and generator, types and applications of GANs.
Marks: 17–18 in the end-semester exam
- Deep generative models
- Boltzmann machines and deep belief networks
- Generative adversarial networks
- Types and applications of GANs
Unit 6: Reinforcement Learning
7 hoursDeep RL, MDPs, the RL framework and its challenges, dynamic programming, Q-learning and deep Q-networks, deep Q recurrent networks, Tic-Tac-Toe.
Marks: 17–18 in the end-semester exam
- Deep reinforcement learning and MDPs
- Dynamic programming for reinforcement learning
- Q-learning and deep Q-networks
Unit contents are summarised from the official SPPU syllabus . AiCanCode.org is not affiliated with Savitribai Phule Pune University.