Browse Teach Curate

Hidden State in Recurrent Networks

A fixed-size vector recomputed each step from the previous state and current input, serving as the network's entire running summary of the past. Its fixed capacity is both the efficiency win and the information bottleneck of recurrence.

jeremy Video

Recurrent Neural Networks (RNNs), Clearly Explained!!!

Explains RNNs for variable-length sequences: feedback loops carrying a hidden state, unrolling the recurrence, shared weights across time steps, and the vanishing/exploding gradient problem that moti…