Using linear programming for Bayesian exploration in Markov decision processes
Why this work is in the frame
A frame that forgets how it found something cannot be audited. These are the routes that admitted this work.
Bibliographic record
Abstract
A key problem in reinforcement learning is finding a good balance between the need to explore the environment and the need to gain rewards by exploiting existing knowledge. Much research has been devoted to this topic, and many of the proposed methods are aimed simply at ensuring that enough samples are gathered to estimate well the value function. In contrast, [Bellman and Kalaba, 1959] proposed constructing a representation in which the states of the original system are paired with knowledge about the current model. Hence, knowledge about the possible Markov models of the environment is represented and maintained explicitly. Unfortunately, this approach is intractable except for bandit problems (where it gives rise to Gittins indices, an optimal exploration method). In this paper, we explore ideas for making this method computationally tractable. We maintain a model of the environment as a Markov Decision Process. We sample finite-length trajectories from the infinite tree using ideas based on sparse sampling. Finding the values of the nodes of this sparse subtree can then be expressed as an optimization problem, which we solve using Linear Programming. We illustrate this approach on a few domains and compare it with other exploration algorithms.
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.
Full frame distilled prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. Learned from the 10,348 direct Codex labels and 10,348 direct Gemma labels. Candidate is the union of thresholded teacher heads; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels or direct frontier model labels.
Codex and Gemma teacher scores by category
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.001 | 0.001 |
| Meta-epidemiology (narrow) | 0.000 | 0.000 |
| Meta-epidemiology (broad) | 0.000 | 0.000 |
| Bibliometrics | 0.001 | 0.001 |
| Science and technology studies | 0.000 | 0.000 |
| Scholarly communication | 0.000 | 0.001 |
| Open science | 0.001 | 0.000 |
| Research integrity | 0.000 | 0.000 |
| Insufficient payload (model declined to judge) | 0.000 | 0.000 |
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 it