ashwin.system60 fps

~/arcade / gradient-golf

Gradient Descent Golf

Every stroke runs θ ← θ − lr·∇loss with a real numerically-computed gradient. Swing too hard and you'll overshoot the ravine; too soft and you'll fossilize in a local minimum. Three holes, par scoring, and the same dial every ML engineer argues with.

GAME 02 · GRADIENT DESCENT GOLF

Putt the loss to the global minimum

the only golf where the ball rolls uphill if you swing too hard
strokes 0HI ---
aim: reach the flag at the global minimum
Each stroke is one optimizer step: θ ← θ − lr·∇loss. Hole 2 has a local-minimum trap; hole 3 is a noisy ravine. Tuning the learning rate mid-round is the entire game — same as training.
more

arcade.siblings

free · client-side · nothing leaves this tabbuilt by Ashwin H — AI engineer
careertime INFO boot: system online  0%