BayBE: A Bayesian Back End for Experimental Planning in the Low-To-No-Data Regime
summary
BayBE is an open‑source Apache‑2 experimental planner that combines Bayesian optimization with BoTorch/GP tools, a probabilistic surrogate, and acquisition functions to recommend experiments in the low‑to‑no‑data regime. It implements chemical encoding for categorical chemicals, batch recommendations, transfer learning, multifidelity and active‑learning (NIPV) policies, and a modular API using attrs, Kus and Hypothesis for testing. Benchmarks on reaction screening show roughly one‑third to one‑quarter the experiments of a CCD DOE to reach the same yield, and it runs in production driving self‑driving labs.
AI-written summary · about
chapters (7)
- 0:00BayBE introduction
- 7:15Bayesian optimization basics
- 17:30BayBE features
- 28:00BayBE code design
- 50:00Class design with AT
- 1:00:10Handling dependencies
- 1:11:19Hands-on demo
related talks
31mCost-effective data annotation with Bayesian experimental design
25mScaling Bayesian Experimentation
46mBayesian Optimization Accelerator
01h 10mBayesian Optimization and Machine Learning for Accelerating Experiments in the Physical Sciences
28mBayesian Optimization- Fundamentals, Implementation, and Practice
01h 03mAn Overview of Probabilistic Programming
39mModern Bayesian Experimental Design
04mInterview