Trustworthy Reinforcement Learning under Constraints and Perturbations

Summary

Starter note seeded from title and cover page. This thesis looks like a broad source for trustworthy RL under explicit constraints and perturbation robustness, and it should likely become an anchor source for your vault.

Key Claims

• Full claims pending deeper ingest.
• Expected role in the wiki: connect constrained RL, robustness, and trustworthiness under changing conditions.

Methods / Formalism

• Constraints and perturbation analysis in RL.
• Thesis-level synthesis likely spanning multiple methods or case studies.

Evidence / Experiments

• Not yet reviewed in detail.

Connections

• Broad source for Safe Reinforcement Learning and Continual Learning.
• Likely useful for connecting robustness, constraints, and trustworthy adaptation.
• Cleanup triage (2026-05-22): keep this as a broad thesis anchor rather than merging it into a single concept. Route constraint material toward Safe Reinforcement Learning and Safe Multi-Agent Reinforcement Learning, and route perturbation/adaptation material toward Continual Learning once chapter-level claims are reviewed.

Open Questions

• Which perturbation models are treated?
• Does the thesis lean more on formal guarantees, robust optimization, or empirical trustworthiness?

Citation

Zhang, L. (2025). Trustworthy Reinforcement Learning under Constraints and Perturbations. PhD thesis, University of Auckland.