Gradient descent

An iterative optimization method that adjusts model parameters in the direction that reduces a loss function, scaled by a learning rate. Too high a learning rate overshoots and can diverge, too low a rate converges so slowly that training looks stalled.

Why exams ask this

Tested as a learning-rate scenario: loss that explodes or oscillates points at a rate too high, loss that barely moves points at a rate too low. The exam wants the learning rate identified as the lever, not the optimizer algorithm itself.

Relevant to

Related concepts

Resources

No resources linked to this concept yet.