Answers are grounded in this paper and cite their sources.
Finding diverse and high-quality candidates in vast design spaces remains an important challenge in scientific research, including molecular design and material discovery. A recent line of work treats observed properties as samples from an unknown distribution and selects candidates that promote more uniform coverage of the property space using Stein discrepancy as the metric. However, these methods do not explicitly account for the information gain of newly acquired data, which can limit exploration of novel regions. In addition, they are not designed for efficient batch selection and may suffer performance degradation when applied in batch settings. Moreover, they incur substantial computational cost due to both large per-iteration overhead and increasing cost as the number of observations grows. To address these limitations, we incorporate predictive 1 uncertainty into the objective function and introduce diversity-aware batch selection strategies to promote informative and diverse candidates. In addition, we propose a sampled Stein novelty estimator and efficient tensor calculations that significantly reduce computational overhead. Performance and runtime evaluations on molecular and materials datasets show that our method enables scalable and sample-efficient expansion of property-space coverage.
No takes yet. Share an insight, caveat, or question.
Fujii et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: