SPPU · Final Year Artificial Intelligence and Data Science · 2020 Course
Deep Learning
417532D · Elective V · Semester VIII · 3 credits · In-semester 30 + end-semester 70
One of the Elective V options.
Unit 1: Foundations of Deep Learning
6 hoursHistory, advantages and challenges; learning representations; architecture principles; hyperparameters: learning rate, regularisation, momentum, sparsity; cost functions and backpropagation; gradient descent; vanishing and exploding gradients; SGD, AdaGrad, RMSProp and Adam.
Marks: 15 in the in-semester exam
- Hyperparameters in deep learning
- Gradient descent variants
- Optimisers: AdaGrad, RMSProp and Adam
- Vanishing and exploding gradients
- Regularisation and dropout
Unit 2: Deep Neural Networks
6 hoursBiological neurons; perceptrons for AND, OR, NOT and XOR; feed-forward networks; forward and backpropagation; linear, sigmoid, tanh, softmax and ReLU; loss functions for regression, classification and reconstruction.
Marks: 15 in the in-semester exam
- Biological and artificial neurons
- Perceptron learning rule
- Multilayer perceptron
- Backpropagation
- Activation functions
- Loss functions
Unit 3: Convolutional Neural Networks
6 hoursCNN structure; padding and strides; ReLU, pooling and fully connected layers; local response normalisation; training CNNs.
Marks: 17–18 in the end-semester exam
- Convolution and feature maps
- Padding, stride and pooling
- Training a CNN
Unit 4: Recurrent Neural Networks
6 hoursUnfolding graphs; bidirectional and deep RNNs; encoder–decoder models; recursive nets; long-term dependencies; echo state networks and leaky units; LSTM and gated RNNs; explicit memory; practical methodology.
Marks: 17–18 in the end-semester exam
- Recurrent neural networks
- Bidirectional and deep RNNs
- Encoder-decoder sequence models
- LSTM
- Echo state networks
Unit 5: Deep Generative Models
8 hoursBoltzmann machines; deep belief networks; GANs, their generator and discriminator, types and applications.
Marks: 17–18 in the end-semester exam
- Boltzmann machines and deep belief networks
- Generative adversarial networks
- Types of GANs
Unit 6: Deep Reinforcement Learning
6 hoursMDPs and the RL framework; dynamic programming; Q-learning and deep Q-networks; deep recurrent Q-networks; RL for tic-tac-toe.
Marks: 17–18 in the end-semester exam
- Deep reinforcement learning
- Deep Q-networks
Unit contents are summarised from the official SPPU syllabus . AiCanCode.org is not affiliated with Savitribai Phule Pune University.