LSTM memory lab

This shows one scalar feature of the cell. The sliders are gate outputs; a trained LSTM computes them from input and previous hidden state.

Choose one step

Previous cell c_t−1 Forget gate f Input gate i Candidate content c̃ Output gate o

Set f=1 and i=0 to carry the old cell unchanged. Set o=0 to hide the cell from the exposed hidden state without erasing it.