Recurrent neural networks

Understands that RNNs maintain hidden state to process sequences; knows LSTM was invented to fix vanishing gradients.

Implements a simple RNN and LSTM in PyTorch for sequence classification or language modelling.

Applies bidirectional LSTMs, stacked RNNs, encoder-decoder with attention for seq2seq tasks.

Analyses BPTT gradient flow; applies advanced tricks (zone-out, variational dropout); understands when to prefer RNN vs Transformer.

Contributes to SSMs (Mamba), linear-complexity sequence models, or time-series deep learning.