gradient-descent #

A ball rolling down a parabolic loss surface, settling at the minimum. Physics simulation with velocity and damping creates natural motion.

canvasclick to interact

⏮◀◀▶▶STEP0.25x1xZOOM

t=0s

practical uses #

01.Explaining optimization fundamentals to beginners
02.Illustrating why learning rate matters (ball momentum)
03.Visualizing convergence behavior in training
04.Demonstrating local vs global minima concepts

technical notes #

Gradient calculated analytically (2x for parabola). Velocity updated with fixed damping factor. Auto-resets when settled to show multiple descent trajectories.

← neural-network binary-tree →