cs.LGAug 7, 2026

Multiscale Reward Hedging from Correct Demonstrations

Authors: Pahan Dewasurendra

Organizations: Johns Hopkins University

Abstract

Learning from correct demonstrations is harder than supervised learning when many answers are correct: after predicting, the learner sees one valid answer but not whether its own answer was valid, nor any reward. Existing reward-hedging guarantees consequently assume a finite reward class. We give the first horizon-free guarantee for continuous classes. The key is to hedge in one shared vote over tolerant optimality tests at every accuracy scale. A target reward has one surviving proxy per scale, and a prediction with gap above that scale doubles the proxy. This yields the simultaneous tail bound ∣{t:ℓt>2−j}∣≤log⁡2N(G,2−j−1)+j|\{t:\ell_t>2^{-j}\}|\leq \log_2\mathcal N(\mathcal G,2^{-j-1})+j, where G\mathcal G is the class of optimality-gap functions. Integrating the tails gives cumulative hidden gap bounded by a metric-entropy integral, independently of the number of rounds. Polynomial entropy (A/ε)d(A/ε)^d gives O(dlog⁡A)O(d\log A) total gap and a fast O(d/m)O(d/m) statistical rate. For bounded linear contextual recommendation, the result is O(d)O(d) regret for arbitrary compact menus. This is the first polynomial finite bound without structural restrictions on the menus, at the price of improper prediction. Although the general vote can be expensive, it is exactly polynomial-time for one-dimensional Lipschitz parameter curves. Fixed-radius rank-two recommendation takes O(KT2)O(KT^2) time for menus of size KK. We also prove an Ω(d)Ω(d) lower bound, low-rank and bounded ReLU-network corollaries, and a robust theorem that adds only the demonstrator's cumulative suboptimality. A reproducible adaptive stress test illustrates the predicted scale adaptation. After factorization, an exact MovieLens audit runs in 1.7 CPU seconds across ten users and improves mean latent gap over both a demonstrated-rating policy and a proper online baseline. The learner uses only action demonstrations and never observes a reward or a loss.

Explore similar work

Sep 22, 2026stat.ML

Tight Regret Bound for Online Inverse Linear Optimization via Multiscale Matrix Weights

We study online inverse linear optimization with a fixed unknown linear utility: in each round, an environment presents a compact action set, the learner recommends an action from it, and the environment returns an action that maximizes the utility over the same set. When the utility vector and the actions lie in the dd-dimensional Euclidean unit ball, we give a randomized algorithm whose regret---the cumulative utility shortfall relative to optimal actions---is O(d)O(\sqrt d) in expectation for every time horizon, without knowledge of the horizon. The dependence on dd is optimal up to a constant factor by the known Ω(d)Ω(\sqrt d) lower bound for horizons T≥dT\ge d. Our algorithm maintains matrix multiplicative weights on polynomial feature spaces at geometrically spaced scales. It selects a recommendation distribution by solving a linear program and updates its score matrices by comparing the available actions with the feedback action. With rational oracle outputs and feedback actions, an implementation computable relative to a linear-optimization oracle preserves the O(d)O(\sqrt d) regret bound. Whether the same rate is attainable with running time polynomial in the dimension, horizon, and input length remains open.
Shinsaku Sakaue
Sep 14, 2026cs.LG

Bandits with Probing: Optimal Regret and the Limits of Winner Feedback

A learner probes at most kk of nn arms each round, receives the maximum of their rewards in [0,1][0,1], and competes with the best fixed arm. When does the probing advantage pay for learning? We determine two minimax laws. Under independent stochastic rewards with winner feedback (the maximum and a winning label), or on arbitrary fixed sequences given a single signed contrast between block maxima, the minimax regret has order Φn,k(T)=min⁡{n−knT,n−kk}Φ_{n,k}(T)=\min\{\frac{n-k}{n}T,\frac{n-k}{k}\}, 2≤k<n2\le k<n. Under winner feedback, both arbitrary joint i.i.d. rewards and fixed sequences have minimax regret of order Rn,k(T)=n−knmin⁡{T,n+Tk,nTk}R_{n,k}(T)=\frac{n-k}{n}\min\{T,\frac{n+T}{k},\sqrt{\frac{nT}{k}}\}. Both laws have universal constants and anytime upper bounds. The first reduces regret to a pure coverage cost: same-round contrasts absorb the stability cost, and independence permits exact resampling whose gains fund sample advancement. The second adds a learning cost that becomes comparable to coverage at horizon nn; beyond nknk, numerical maxima improve over labels alone. The lower bound allows every adaptive action size.
Yongjie Guan
May 29, 2026cs.LG

Inverse Reinforcement Learning without an Optimal Demonstrator: A Feasible Reward Set Approach

Inverse reinforcement learning (IRL) typically assumes demonstrations from a single optimal demonstrator, but in many applications data come from multiple imperfect demonstrators with heterogeneous suboptimality levels. We study reward learning in this setting through a feasible-reward-set framework: for each demonstrator, we encode its declared suboptimality level as a linear constraint and intersect the resulting feasible sets across demonstrators. Our theoretical analysis shows that the joint feasible set shrinks monotonically as data are added, and we give an exact characterization of when a new demonstrator strictly tightens it. We further establish two recovery guarantees for the feasible reward set of the ground-truth optimal demonstrator: one bound depends on closeness to the optimal occupancy, while the other requires only sufficient coverage and no near-optimal demonstrator. On the practical side, we introduce strategies to address the inherent reward ambiguity in the obtained reward set and provide an offline algorithm with function approximation for high-dimensional environments. Experiments in tabular grid-world and large language model (LLM) fine-tuning settings are consistent with the theoretical predictions and demonstrate the effectiveness of the proposed framework over baselines.
Kihyun Kim, Shripad Deshmukh, Nikos Vlassis +1