bayesian-optimization

Animated closed-loop Bayesian Optimization: a hidden objective f(x) (faint), a probabilistic surrogate summarized by μ(x) and an uncertainty band (σ), and an acquisition function a(x) that trades exploitation (-μ) and exploration (κσ). Each cycle highlights x_next = argmax a(x), evaluates f(x_next) (falling sample), then updates the surrogate.

canvasclick to interact

t=0s

practical uses

01.Hyperparameter tuning of ML models with expensive training runs
02.Optimizing engineering design parameters using costly simulations (CFD/FEA)
03.Calibrating experiments or A/B settings when each evaluation is expensive

technical notes

Self-contained canvas 2D animation with a simple RBF-kernel surrogate (kernel regression + heuristic uncertainty) and UCB acquisition. Time is segmented into phases (update → acquire → evaluate) over a 3s loop. Blocky grid-snapped rendering with green-on-black palette.

← activation-functions eigenvalues-and-eigenvectors →