Data Science: A Physics Perspective
Gradient descent as a ball rolling in a loss basin: fit a plane to 3D data with momentum, watch stability limits, conditioning and live numerical checks, in a Blender-built Three.js lab.
Linear regression trained by gradient descent with momentum, seen as a steel ball rolling down a loss basin next to the data cloud it is fitting. The step size is measured against the exact stability limit, the correlation between features stretches the valley, and seven independent numerical checks confirm the maths live.
Gradient descent as a heavy ball rolling in a loss basin: fit a plane to 3D data with momentum, see the exact stability limit, the effect of correlated features, and seven live numerical checks, in a Blender-built Three.js lab.
3D · Three.js / WebGL renderer · 60 FPS target · runs fully client-side, no install