Learning a Ranking from Human Feedback in Log-Concave Random Utility Models
In both settings, we model human feedback using a random utility model with log-concave noise and study the number of observations needed to recover an $ε$-accurate ranking with high probability.
ProofPaper ↗
Key points
- We study the problem of recovering the ranking of a fixed set of items according to their unknown numerical utilities.
- Under full-ranking feedback, each interaction reveals a noisy ranking of all items, whereas under winner-only feedback, it reveals only the item ranked first.
- For both feedback types, we establish worst-case sample-complexity lower bounds and develop algorithms that match these bounds up to logarithmic factors.
- Our results show that the ranking problem under winner-only feedback is intrinsically harder by exposing the sample complexity dependence on the minimum winning probability across the item set.
Sources (1)
- [1]Learning a Ranking from Human Feedback in Log-Concave Random Utility ModelsarXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 6, 08:41 AM
In both settings, we model human feedback using a random utility model with log-concave noise and study the number of observations needed to recover an $ε$-accurate ranking with high probability.
We study the problem of recovering the ranking of a fixed set of items according to their unknown numerical utilities.
Extractive summary: sentences quoted from the sources.
