DyRAMO

Dyramo is a framework for exploring promising candidates (e.g., molecules) by performing multi-objective optimization with a generative model while accounting for the reliability (applicability domain) of multiple property-prediction models, thereby avoiding reward hacking.