Deep Learning · Sequences

RNN Memory

Drag Wₕ — the hidden state carries the past forward. High Wₕ remembers; low Wₕ forgets.

recurrent Wₕ 0.70 input
hₜ xₜ
Wₕ
0.70
memory
regime
short memory
hidden hₜinput xₜ
tip: send one input pulse, then raise Wₕ and watch the echo last longer
rnn.py