Weight

Also known as · weights

A number controlling how much one input matters to a neuron; it is learned in training. All weights start as random numbers.

When to use

Used in the weighted sum z=w1x1+w2x2+⋯+bz = w_1 x_1 + w_2 x_2 + \dots + b in every forward-pass calculation (e.g. weights 0.5, 0.8 and 0.1 on hour, rain and events). Training adjusts each weight by Gradient Descent, with Backpropagation tracing how much each contributed to the error.

See DS Lec 8.

Appears in

No references yet.