SPPU · Third Year Artificial Intelligence and Data Science · 2024 Pattern
Deep Learning
PCC-352-AID · Semester VI · 4 credits · CCE 30 + end-semester 70
Unit 1: Artificial Neural Networks
9 hoursBiological and artificial neurons; activation functions; single- and multi-layer feed-forward networks; supervised and unsupervised learning, training, testing and loss; the perceptron and its learning rule and limits; the multilayer perceptron, forward propagation and backpropagation.
Marks: 14 in the end-semester exam · CCE: Unit test (12 marks, Units 1 & 2)
- Biological and artificial neurons
- Feed-forward network architectures
- Perceptron learning rule
- Multilayer perceptron
Unit 2: Training Deep Networks
9 hoursDeep networks and how they differ from ML; ReLU, sigmoid, tanh and softmax; forward propagation; MSE and cross-entropy; backpropagation; batch, stochastic and mini-batch gradient descent; SGD, AdaGrad, RMSProp and Adam; vanishing and exploding gradients; weight initialisation; L1, L2 and dropout.
Marks: 14 in the end-semester exam · CCE: Unit test (12 marks, Units 1 & 2)
- Activation functions
- Loss functions
- Backpropagation
- Gradient descent variants
- Optimisers: AdaGrad, RMSProp and Adam
- Vanishing and exploding gradients
- Regularisation and dropout
Unit 3: Convolutional Neural Networks
9 hoursConvolution, kernels and feature maps; filter size, padding and stride; pooling; CNN design; LeNet, AlexNet, VGG and ResNet; transfer learning and fine-tuning; YOLO.
Marks: 14 in the end-semester exam · CCE: Assignments or case study (12 marks, Units 3 & 4)
- Convolution and feature maps
- Padding, stride and pooling
- CNN architectures: LeNet to ResNet
- Transfer learning
- Object detection with YOLO
Unit 4: Recurrent Neural Networks
9 hoursSequence models; RNNs, bidirectional and deep RNNs; LSTM and GRU; introduction to transformers; Word2Vec and GloVe; uses in NLP, speech, time series, video and health signals.
Marks: 14 in the end-semester exam · CCE: Assignments or case study (12 marks, Units 3 & 4)
- Recurrent neural networks
- LSTM
- GRU
- Word embeddings: Word2Vec and GloVe
Unit 5: Generative Models and Deployment
9 hoursGANs; autoencoders and variational autoencoders; model compression and pruning; deployment with REST APIs and TensorFlow Lite; edge AI; ethics and energy efficiency.
Marks: 14 in the end-semester exam · CCE: Seminar, open-book test or quiz (6 marks, Unit 5)
- Generative adversarial networks
- Autoencoders and VAEs
- Model compression and pruning
- Deploying deep learning models
Unit contents are summarised from the official SPPU syllabus . AiCanCode.org is not affiliated with Savitribai Phule Pune University.