Reproduce “Learning the Value Systems of Agents with Preference-based and Inverse Reinforcement Learning” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Andrés Holgado-Sánchez, Holger Billhardt, Alberto Fernández, Sascha Ossowski
The authors separate value grounding from value-system inference. They learn each value’s reward from quantified trajectory comparisons, then use maximum-entropy inverse reinforcement learning to infer linear weights from behavior. In two synthetic environments, groundings preserved preference orderings with at least 98% accuracy, and Firefighters visitation-count error stayed below 0.005; this does not validate real-world values or uniquely identify weights.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.