Mapping of COs with PSOs and POs: Page 157 of 241 IV Recurrent Neural Networks 13 20 16 Computational graphs 1 17 RNN design 1 18 Encoder – decoder sequence to sequence architectures 2 19 Deep recurrent networks 2 20 Recursive neural networks 3 21 Modern RNNs LSTM and GRU. 2 22 Practical use cases for RNNs. 2 V Open Ended Module – Applications 12 1 Implement the tasks from the following: • Implement and analyze single-layer and multi-layer perceptrons with activation functions, loss functions, and backpropagation. • Investigate overfitting, vanishing gradients, and exploding gradients in deep networks, and explore solutions.
• Design and train a CNN for image classification, focusing on convolution, pooling, and efficient algorithms. • Implement and train RNNs, including LSTM and GRU, for sequence prediction, and analyze their performance. • Train a deep feed-forward network on a complex dataset, using eigenvalues and eigenvectors, and compare optimization techniques. 12 PSO1 PSO2 PSO3 PSO4 PSO5 PSO6 PO1 PO2 PO3 PO4 PO5 PO6 CO 1 2 1 2 2 2 1 CO 2 2 1 3 2 2 2 CO 3 3 1 3 2 3 2 CO 4 3 2 3 3 3 3 CO 5 3 2 3 3 3 3 CO 6 3 2 3 3 3 3