The interactive film is in the book's Companion

Play it, pause it, turn it and drag it with Mathematics for AI and Machine Learning.

Get the book

Sophia: Newton Steps with a Clip

Chapter 12 · Mathematics for AI and Machine Learning · Advanced · 2:12

An interactive live film: on a valley with a negatively curved plateau, gradient descent bounces, Adam swings, Newton walks uphill, and Sophia divides by the diagonal Hessian with a clip; the reader drags the start and sets the clip.

Narration

Here is a loss with two very different directions. Across the valley, along y, the walls are steep. Along x, the floor is a gentle well, and far from the centre it flattens into a plateau that even curves slightly downward.

Gradient descent, in red, uses one step size for every direction. A step large enough to move on the plateau makes it bounce from wall to wall across the valley, and it barely crawls along the floor.

The rest of the narration is in the book's Companion.

More from Chapter 12

Interactive1:08Gradient Descent in an Ill-Conditioned ValleyChapter 12Preview1:28The Descent Lemma: A Bowl That Bounds Every StepChapter 12Preview1:33Heavy-Ball Momentum: Zig-Zags Cancel, the Valley Floor Adds UpChapter 12Preview1:43A Large Step Selects a Flatter MinimumChapter 12Preview2:14Adam and AdamW: A Step Size for Every CoordinateChapter 12Preview1:55Backtracking Line Search and the Armijo ConditionChapter 12Preview1:50Proximal Gradient and Soft-Thresholding: Exact ZerosChapter 12Preview1:53KKT Conditions: the Saddle Point of the LagrangianChapter 12Preview1:52Mirror Descent Stays Inside the SimplexChapter 12Preview1:55The Polyak-Łojasiewicz ConditionChapter 12