Home▸AI & Machine Learning▸Inside an LSTM Cell

🧠 Inside an LSTM Cell

A 3D walkthrough of an LSTM cell's memory highway, showing how the forget, input and output gates decide what a network keeps, updates and reveals at each time step.

AI & Machine Learning3DAdvanced60 FPS

A 3D memory highway shows how an LSTM cell's forget, input and output gates decide what to keep, what to write in, and what to reveal at each time step — illustrated with a toy airline-passenger forecast.

🔬 What It Demonstrates

The cell state flows left to right along a lit highway. A forget-gate valve scales down old memory, an input-gate valve scales in a new candidate value, and an output-gate valve squashes and releases part of the result as the hidden state that drives the forecast panel.

🎮 How to Use

Drag the forget, input, candidate and output sliders to see the bar banks and forecast line respond instantly, or flip on Auto time-step to watch the gates oscillate through a short sequence on their own.

💡 Did You Know?

Because the cell-state update is additive (old memory plus a gated new term) rather than a full rewrite, gradients can flow through many time steps without vanishing — the core trick that made LSTMs so effective for long time-series like passenger demand.

⚙ Under the hood

A 3D walkthrough of an LSTM cell's memory highway, showing how the forget, input and output gates decide what a network keeps, updates and reveals at each time step.

artificial intelligenceneural networkslstmdeep learningrecurrent networksmachine learningThree.js

3D · Three.js / WebGL renderer · 60 FPS target · runs fully client-side, no install