Landscape
Optimizer
Learning rate
Gradient noise
Steps / frame
landscape = Bowl
optimizer = SGD
step = 0
loss = —
|∇f| = —
α = 0.1