Gradient descent finally has a lab session.
Predict first, run the experiment, then rebuild the idea from evidence when the bench proves you wrong.
One guided lesson · 20 min
Optimization Canyon
Explore why a learning rate that works on one loss surface can zigzag, stall, or diverge on another.