SEQUENCES · GATES · MEMORY LSTM, Keep something. Change something. Decide what to reveal. Follow a memory through five short chapters.
Start reading → Cell storage and exposed output Prescribed-gate illustration with f0.9, counter-cue write0, output0.5. The cell retains a decaying positive cue while hidden output is a smaller readout. The lab explains and lets you vary these operations. A MEMORY, CARRIED FORWARD 1 2 3 4 5 6 cell c hidden h sequence step One early cue, two state traces. A prescribed-gate example—not a trained model.
Long short-term memory, from one update to a sequence. Start with basic ML and arithmetic; we introduce the notation as we go. No model download or prior recurrent-network knowledge required.
01 Remembering a sequence A later decision may depend on an earlier event. A recurrent state carries something forward.
↗ 02 Inside the cell Keep some old state, add a proposed write, then decide how much to expose.
↗ 03 Gates over time Keep the six events fixed. Change one multiplier and trace the consequence.
↗ 04 A path for gradients The carry route can preserve a learning signal. It does not guarantee unlimited memory or easy training.
↗ 05 From cell to model Learned weights choose the gates. A separate task readout turns hidden features into a prediction.
↗