Home▸AI & Machine Learning▸Hyperparameter Optimization Landscape

🎯 Hyperparameter Optimization Landscape

A 3D loss-surface explorer comparing Grid Search, Random Search and Bayesian Optimization as they hunt for the best hyperparameters.

AI & Machine Learning3DAdvanced60 FPS

A 3D loss surface shaped by two hyperparameters — a log-scale learning rate and a log-scale regularization strength — that Grid Search, Random Search and Bayesian Optimization each try to minimize with a limited trial budget.

🔬 What It Demonstrates

Height and color encode validation loss across the two-hyperparameter space. Watch how each search strategy places its trials — a rigid lattice, a random scatter, or an informed sequence that homes in on the deepest valley — and compare how quickly each finds the true minimum.

🎮 How to Use

Pick a search strategy, set the trial budget and how rugged (multimodal) the loss surface is, then watch trials appear live. The pulsing indigo sphere marks the best hyperparameter combination found so far; drag to rotate and scroll to zoom.

💡 Did You Know?

Bergstra & Bengio's 2012 paper "Random Search for Hyper-Parameter Optimization" showed random search often beats grid search for the same compute budget, because real loss surfaces are usually dominated by only one or two truly important hyperparameters.

⚙ Under the hood

A 3D loss-surface explorer comparing Grid Search, Random Search and Bayesian Optimization as they hunt for the best hyperparameters.

machine learningoptimizationhyperparametersgrid searchrandom searchbayesian optimizationloss functionThree.js

3D · Three.js / WebGL renderer · 60 FPS target · runs fully client-side, no install