Bayesian Optimization

Surrogate models, acquisition functions, and closing the BO loop — for efficient molecular and materials discovery with few experiments.

Courses

3 resources

Exploring Bayesian Optimization — Distill.pub

Interactive article with explorable animations of GP surrogates, acquisition functions (EI, UCB, PI), and the full BO loop. The most accessible introduction.

Distill.pub 2020 · Interactive

Gaussian Processes for Machine Learning — Rasmussen & Williams

The standard reference for Gaussian processes. Free PDF from the authors. Chapters 1–3 cover GP regression and classification.

MIT Press · Free PDF

A Tutorial on Bayesian Optimization — Frazier (2018)

Accessible tutorial covering GP surrogates, acquisition functions, and convergence guarantees. The go-to reference for BO theory.

arXiv:1807.02811 · Free

Tutorials

2 resources

BoTorch Tutorials

Official Colab notebooks covering single-objective BO, constrained BO, multi-objective BO, and integration with Ax for experiment management.

BoTorch.org · Colab notebooks

End-to-end tutorials for adaptive experimentation using Ax: parameter spaces, trial tracking, early stopping, and multi-fidelity optimisation.

Papers

4 resources

Shahriari et al. — "Taking the Human Out of the Loop: A Review of Bayesian Optimization" (2016)

The standard BO review article covering surrogates, acquisition functions, and applications.

Proc. IEEE 2016

Griffiths & Hernández-Lobato — "Constrained Bayesian Optimization for Automatic Chemical Design" (2020)

BO in the latent space of a molecular VAE with property constraints.

ACS Central Science 2020

Daulton et al. — "Differentiable Expected Hypervolume Improvement for Parallel Multi-Objective Bayesian Optimization" (2020)

Multi-objective BO via differentiable acquisition functions for Pareto-optimal molecular candidates.

NeurIPS 2020 · arXiv:2006.05078

Hase et al. — "Phoenics: A Bayesian Optimizer for Chemistry" (2018)

BO framework designed for chemical reaction and materials optimisation, handling categorical and continuous variables.

ACS Central Science 2018

Tools

4 resources

PyTorch-based Bayesian optimisation library from Meta. Supports GPs, multi-objective BO, constrained BO, and qNEI.

Adaptive experimentation platform built on BoTorch. Handles parameter spaces, trial tracking, and early stopping.

Scalable GP implementation in PyTorch used by BoTorch internally. Supports exact and approximate inference for large datasets.

Interactive BO code generator: configure your optimisation problem in a GUI and export ready-to-run BoTorch/Ax code.

Datasets

3 resources

Buchwald-Hartwig Reactions (Doyle Lab)

High-throughput Buchwald-Hartwig cross-coupling yields across 3955 reactions, commonly used to benchmark BO and yield prediction models.

Doyle Lab · Free

Perovskite Bandgap Dataset (Citrination)

Experimental and computed bandgap data for perovskite compositions. Frequently used as a real-world BO benchmark for materials optimisation.

Citrine Informatics · Free

Hartmann Benchmark Functions (BoTorch)

Synthetic benchmark functions (Hartmann-3, Hartmann-6, Branin) built into BoTorch for testing and comparing BO algorithms.

BoTorch · Built-in