▚ surface · see it

Gradient Descent

How a model learns by following the slope of its own error.

01 · gradient-direction

gradient-direction

A gradient always points uphill.

02 · learning-rate

learning-rate

The learning rate controls how big each step is.

▚▚ deeper · understand it

Gradient Descent

01 · gradient-direction

gradient-direction

∇f(x) is the direction of steepest ascent. Descent flips its sign.

02 · learning-rate

learning-rate

Too large and you overshoot the minimum. Too small and training takes forever.

▚▚▚ deepest · build it

Gradient Descent

01 · gradient-direction

gradient-direction

x = x - lr * grad(f, x) # one step of gradient descent
02 · learning-rate

learning-rate

lr = 0.01 # typical starting value; tune by watching the loss curve
gradient-direction
learning-rate
] [ dive · rise1 2 3 level