Momentum

Optimization · Formula · Appears in 1 lesson

vt+1=βvt+∇θL,θt+1=θt−ηvt+1v_{t+1} = \beta v_t + \nabla_{\theta} L, \qquad \theta_{t+1} = \theta_t - \eta v_{t+1}

Keeps a running average of past gradients so the step has inertia. Consistent directions build up speed while oscillating ones cancel out, which is what gets you along a narrow valley instead of bouncing across it.

Symbols

vv
The velocity, an exponentially weighted average of gradients.
β\beta
The decay rate, usually around 0.9.
Esc
↑↓ navigate↵ open? shortcuts

Keyboard shortcuts

Press Esc to close

Anywhere

⌘K
Search and commands
/
Search
?
This sheet
Esc
Close

Reading a lesson

J
Next lesson
K
Previous lesson
B
Bookmark