These notes cover architectures designed for sequential and temporal data:
- Recurrent Neural Networks: Hidden state propagation, unrolling through time, and backpropagation through time (BPTT).
- Vanishing & Exploding Gradients: Why vanilla RNNs struggle with long sequences and how gating solves it.
- LSTM: Forget gate, input gate, output gate, cell state — the full gating mechanism explained visually.
- GRU: Simplified gating with reset and update gates, and comparison with LSTM.
- Bidirectional RNNs: Processing sequences in both directions for richer context.