5 accepted papers
Uniform-reward reinforcement learning from human feedback (RLHF), which trains a single reward model to represent the preferences of all annotators, fails to capture the diversity of opinions across sub-populations, inadvertently favoring dominant groups. The state-of-the-art, MaxMin-RLHF, addresses
While test-time scaling with verification has shown promise in improving the performance of large language models (LLMs), role of the verifier and its imperfections remain underexplored. The effect of verification manifests through interactions of three quantities: (i) the generator’s *coverage*, (i…
This paper introduces a general framework for risk-sensitive bandits that integrates the notions of risk-sensitive objectives by adopting a rich class of {\em distortion riskmetrics}. The introduced framework subsumes the various existing risk-sensitive models. An important and hitherto unknown obse…
The paper considers the problem of estimating a covariate parameter shared by multiple statistical models. Under the objective of estimating the parameter with target reliability with the fewest number of samples from these models, a fundamental question is how to glean samples from the statistical…
This paper investigates the problem of best arm identification in {\sl contaminated} stochastic multi-arm bandits. In this setting, the rewards obtained from any arm are replaced by samples from an adversarial model with probability $\varepsilon$. A fixed confidence (infinite-horizon) setting is con…