Problem 5 - Olympiad

During gradient descent, if the learning rate is too large, which symptom is most probable?

A. Training loss decreases very slowlyB. Model converges to a sharp minimumC. Loss oscillates or explodesD. Weights approach exactly zero

Correct: C

An oversized learning rate causes parameter updates to overshoot the minimum, making the loss bounce or grow exponentially.