RTMNU · Final Year B.Tech Computer Science & Engineering · CBCS
Deep Learning
BTECHCSE702T · Elective IV · Semester VII · 3 credits · College 30 + university 70
One of the Elective IV options.
Unit 1: Basics of Deep Learning
8 hoursThe history of deep learning, the McCulloch–Pitts neuron and thresholding logic; perceptrons, the perceptron learning algorithm and its convergence; multilayer perceptrons and their representation power, sigmoid neurons and feed-forward networks.
Examined in the 70-mark end-semester paper
- History, advantages and challenges of deep learning
- McCulloch–Pitts neuron and thresholding logic
- Perceptron and feed-forward networks
- Perceptron and multilayer perceptron
Unit 2: Training Feed-forward Neural Networks
7 hoursThe representation power of feed-forward networks and how they are trained: gradient descent and its momentum-based, Nesterov accelerated and stochastic forms, AdaGrad, RMSProp and Adam.
Examined in the 70-mark end-semester paper
- Forward and backpropagation
- Gradient descent and vanishing and exploding gradients
- Optimisers: momentum, Nesterov, AdaGrad, RMSProp and Adam
Unit 3: Optimisation, Activation, Initialisation and Regularisation
7 hoursThe optimisation algorithms in practice; activation functions (sigmoid, tanh, ReLU) and Xavier and He initialisation; regularisation — bias and variance, overfitting, hyperparameter tuning, L1 and L2 regularisation, data augmentation, early stopping, and parameter sharing and tying.
Examined in the 70-mark end-semester paper
- Activation functions in deep networks
- Weight initialisation and greedy layer-wise pre-training
- Regression basics and the bias–variance trade-off
- Regularisation: dropout, batch normalisation and early stopping
- Hyperparameters: learning rate, momentum and regularisation
- Transfer learning and data augmentation
Unit 4: Convolutional Neural Networks
7 hoursThe convolution operation and pooling; the LeNet, AlexNet, ZF-Net, VGGNet, GoogLeNet and ResNet architectures; visualising convolutional networks and guided backpropagation.
Examined in the 70-mark end-semester paper
- CNN architecture
- Convolution, padding and strides
- Pooling and fully connected layers
- LeNet and AlexNet
- CNN architectures: VGG, GoogLeNet, ResNet and DenseNet
- Visualising and fooling CNNs: guided backpropagation and Deep Dream
Unit 5: Recurrent Neural Networks
7 hoursRecurrent neural networks, backpropagation through time, vanishing and exploding gradients, long short-term memory cells and gated recurrent units.
Examined in the 70-mark end-semester paper
- Recurrent neural networks
- Backpropagation through time
- Long-term dependencies, LSTM and GRU
Unit contents are summarised from the official RTMNU syllabus . AiCanCode.org is not affiliated with Rashtrasant Tukadoji Maharaj Nagpur University (RTMNU).