Raghuram Shankar
← Bayesian Optimization

Part 2 · Intuition

Does the model really help? Run the experiment

The regret figure in the Background was one function. Here you can run fresh head-to-head episodes yourself: each one draws a new random smooth function, gives EI-driven BO and uniform random search 12 evaluations each, and records who found the higher value.

no episodes yet
Every episode is a real 12-evaluation contest on a fresh random function (both methods run live in your browser). The tally converges to a strong-but-imperfect edge: sample efficiency is a statistical advantage, not a guarantee.