Gradient descent visualizer.
Follow an optimizer across a loss landscape. Compare gradient descent, Momentum, Nesterov, and Adam.
Your experiment starts here
Ready to explore?
Runs in your browser.
Quick start
Get hands-on.
Choose a loss function and optimizer. Click the landscape to set a start, then play or step through the trajectory. Tune the learning rate and momentum.
What to look for
- Analytical gradients for quadratic, Rosenbrock, saddle, and Rastrigin functions.
- Momentum, lookahead updates, and Adam’s bias-corrected moments.
- Loss curves, contour lines, and the effect of noisy gradients.