During training, a model reports a gradient vector
which points in the direction the error grows fastest. To reduce the error, the optimiser must take a step that points in exactly the opposite direction to , with a magnitude of exactly .
What is ?
Drag v, then slide k. Multiplying a vector by a number stretches, shrinks, or flips it — but never changes the line it sits on.
Once you get it right, the reasoning behind the answer shows up here.
Select all that apply.