Bibliographic record
Abstract
In reinforcement learning, the reward that guides the learning process of a decision-maker acting in an environment to achieve a goal can be non-trivial to specify. As such, inverse reinforcement learning (IRL) seeks to observe an expert and explain behaviour in terms of a reward. However, behaviour of humans is linked to the choice of discount factor and arbitrary choices of discount factor in IRL can lead to different rewards and optimal behaviours. Thus, discount factor estimation as part of the IRL framework is crucial to mitigate such bias in the learning process. This thesis evolves a utility-based objective within the maximum entropy IRL algorithmic framework to simultaneously estimate the discount factor and reward in a gradient-based manner via explicit computations of derivatives. The case of a single expert is first investigated prior to extension to multiple experts, in which Expectation-Maximization procedure is employed to learn each expert's cluster of trajectories and obtain maximum likelihood estimates of interest. Experimental and numerical studies on MDP environments, e.g., Grid-World are carried out with evaluative performance measures. Furthermore, an application to real-world traffic data is presented. Driving behaviour of humans may react sub-optimally to congestion leading to loss in traffic capacity. Therefore, a comparison of the observed traffic flow with theoretical maximum flow gives a quantified loss. Recommendations are also made on how a human driver should be nudged to minimize the loss. This is possible by solving three optimization problems: IRL given traffic data to learn actual reward and discount factor, forward reinforcement learning using a flow-based reward to generate ideal driving behaviour leading to theoretical maximum traffic flow, and a least-square fitting procedure to compute ideal rewards. Empirical evaluations on highway 401 in Greater Toronto Area are carried out to validate the approach.
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 | 0.004 | 0.014 |
| Meta-epidemiology (narrow) | 0.001 | 0.001 |
| Meta-epidemiology (broad) | 0.002 | 0.001 |
| Bibliometrics | 0.001 | 0.000 |
| Science and technology studies | 0.000 | 0.002 |
| Scholarly communication | 0.002 | 0.002 |
| Open science | 0.002 | 0.002 |
| Research integrity | 0.001 | 0.003 |
| Insufficient payload (model declined to judge) | 0.002 | 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 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".