🏎️ LR & Optimizer Race Visualizer

Same landscape, same start — push the Learning Rate and see who survives.

📋 How To Use This Lab

Press ▶ Auto Run — all optimizers race the SAME landscape from the SAME point, live.
Drag Learning Rate WHILE it's racing — push it high and watch SGD bounce off the track while Adam stays controlled.
Switch Terrain to "Saddle Point" — watch SGD crawl through the flat pinch while adaptive methods glide through.
Flip Show Momentum Vectors — SGD's arrow only shows the current slope; Adam's blends in its remembered direction too.

Step 0

SGD: loss=—

RMSprop: loss=—

Adam: loss=—

Learning Rate: 0.50

Terrain

SGD + RMSprop + Adam

+ Momentum

Vectors Off

Show Momentum Vectors

🏎️ PACERPress Auto Run and watch all four take the same starting line — then push the Learning Rate and see who keeps control.

🧑‍🏫 PROF. TORCHSGD's step size scales directly with the raw gradient — no safety net. Adam and RMSprop divide by their own gradient history, which naturally keeps them in check.