Coactive Learning Optimizes Autonomous Materials Discovery Pathways
A novel technique known as Coactive learning merges cost-sensitive Bayesian hypothesis discrimination with Gaussian-process Bayesian optimization to determine the optimal recovery pathway in autonomous materials discovery labs. This method tackles a sequential decision-making challenge, featuring a discrete stage for pathway identification and a continuous stage for optimization within that pathway, all while considering varying experimental costs. Given specific assumptions, the anticipated expenditure for a single fixed-budget campaign is limited to the expected costs of pathway identification plus the defined budget for optimization. The technique was tested against synthetic benchmarks informed by selected outcomes from PNNL's CICERO selective-precipitation study (Ritchhart et al., 2026) and showed performance on par with an oracle-pathway reference.
Key facts
- Coactive learning combines EC2-based cost-sensitive hypothesis discrimination with Gaussian-process Bayesian optimization.
- The method addresses sequential decision problems with discrete pathway identification and continuous optimization stages.
- Expected spend is bounded by pathway-identification cost plus capped optimization budget.
- Evaluated on synthetic benchmarks from PNNL's CICERO selective-precipitation study (Ritchhart et al., 2026).
- Performance comparable to an oracle-pathway baseline.
Entities
Institutions
- Pacific Northwest National Laboratory (PNNL)
- CICERO