Too big, it explodes. Too small, it crawls.
The learning rate, step by step: why one number decides whether training works at all. Backpropagation tells every weight which way to move: its gradient. It doesn't say how far. That's the job of the learning rate, written η (eta). Every step of gradient descent is w ← w − η ∇L The gradient comes…
Read the full story at DEV Community — Machine Learning ↗
Timeline · 1 report
- 2026-11-01 00:20 · DEV Community — Machine Learning
Too big, it explodes. Too small, it crawls.