Research
121d agoDistill-Belief Solves Reward Hacking In Source Localization
A new teacher-student framework called Distill-Belief decouples Bayesian correctness from computational efficiency. It uses a particle-filter teacher to provide dense information-gain signals, preventing a student model from exploiting approximation errors. This approach stops reward hacking during inverse source localization.
The Signal
Practitioners can now deploy faster belief models without sacrificing uncertainty estimation accuracy.