Regret-based Reward Elicitation for Markov Decision Processes
Bibliographic record
Abstract
Markov decision processes (MDPs) have proven to be a useful model for sequential decision- theoretic reasoning under uncertainty, yet they require the specification of a reward function that can require sophisticated human judgement to assess relevant tradeoffs. This dissertation casts the problem of specifying rewards as one of preference elicitation and aims to minimize the degree of precision with which a reward function must be specified while still allowing optimal or near-optimal policies to be produced. We demonstrate how robust policies can be computed for MDPs given only partial reward information using the minimax regret criterion. Minimax regret offers an intuitive bound on loss; however, it is computationally intractable in general. This work develops techniques for exploiting MDP structure to allow for offline precomputation that enables efficient online minimax regret computation. To complement this exact approach we develop several general approximations that offer both upper and lower bounds on minimax regret. We further show how approximations can be improved online during the elicitation procedure to balance accuracy and efficiency. To effectively reduce regret, we investigate a spectrum of elicitation approaches that range from the computationally-demanding optimal selection of complex queries about full MDP policies (which are informative, but, we believe, cognitively difficult) to the heuristic selection of simple queries that focus on a small set of reward parameters. Results are demonstrated on MDPs drawn from the domains of assistive technology and autonomic computing. Finally we demonstrate our framework on a realistic website optimization domain, per- forming elicitation on websites with tens of thousands of webpages. We show that minimax regret can be efficiently computed, and develop informative and cognitively reasonable queries that quickly lower minimax regret, producing policies that offer significant improvement in the design of the underlying websites.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
How this classification was reachedexpand
Full frame machine prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. The Gemma side is a direct model label for every work in the frame, read from the title-only record. The Codex side is a classifier learned from the 10,348 direct Codex labels and calibrated to design-weighted sample rates; fields without enough sample support carry no Codex call. Candidate is the union of the two sides; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels.
Distilled classifier scores by category (both heads)
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.009 | 0.038 |
| Meta-epidemiology (narrow) | 0.002 | 0.001 |
| Meta-epidemiology (broad) | 0.002 | 0.002 |
| Bibliometrics | 0.001 | 0.001 |
| Science and technology studies | 0.001 | 0.002 |
| Scholarly communication | 0.002 | 0.003 |
| Open science | 0.002 | 0.002 |
| Research integrity | 0.002 | 0.004 |
| Insufficient payload (model declined to judge) | 0.005 | 0.001 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from itClassification
machine, unvalidatedMachine predicted; a candidate call from one source (direct Gemma or distilled Codex), not a consensus.
How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".