Learning from Delayed Semi-Bandit Feedback under Strong Fairness Guarantees
Bibliographic record
Abstract
Multi-armed bandit frameworks, including combinatorial semi-bandits and sleeping bandits, are commonly employed to model problems in communication networks and other engineering domains. In such problems, feedback to the learning agent is often delayed (e.g. communication delays in a wireless network or conversion delays in online advertising). Moreover, arms in a bandit problem often represent entities required to be treated fairly, i.e. the arms should be played at least a required fraction of the time. In contrast to the previously studied asymptotic fairness, many real-time systems require such fairness guarantees to hold even in the short-term (e.g. ensuring the credibility of information flows in an industrial Internet of Things (IoT) system). To that end, we develop the Learning with Delays under Fairness (LDF) algorithm to solve combinatorial semi-bandit problems with sleeping arms and delayed feedback, which we prove guarantees strong (short-term) fairness. While previous theoretical work on bandit problems with delayed feedback typically derive instance-dependent regret bounds, this approach proves to be challenging when simultaneously considering fairness. We instead derive a novel instance-independent regret bound in this setting which agrees with state-of-the-art bounds. We verify our theoretical results with extensive simulations using both synthetic and real-world datasets.
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 | 0.007 | 0.025 |
| Meta-epidemiology (narrow) | 0.002 | 0.001 |
| Meta-epidemiology (broad) | 0.003 | 0.001 |
| 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 | 0.003 |
| Insufficient payload (model declined to judge) | 0.003 | 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".