Best-of-Tails: Bridging Optimism and Pessimism in Inference-Time Alignment

• URL: http://arxiv.org/abs/2603.06797v1
• Date: Fri, 06 Mar 2026 19:00:54 GMT
• Title: Best-of-Tails: Bridging Optimism and Pessimism in Inference-Time Alignment
• Abstract summary: Inference-time alignment effectively steers large language models by generating multiple candidates from a reference model and selecting among them with an imperfect reward model.<n>We show that the optimal strategy depends critically on the tail behavior of the reward distribution.<n>We introduce Best-of-Tails (BoT), an adaptive inference-time alignment framework that uses Tsallis divergence as a tunable regularizer.
• Score: 20.579613395916024
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Inference-time alignment effectively steers large language models (LLMs) by generating multiple candidates from a reference model and selecting among them with an imperfect reward model. However, current strategies face a fundamental dilemma: ``optimistic'' approaches like Best-of-$N$ suffer from reward hacking, while ``pessimistic'' regularized methods often stifle the exploration needed to discover high-quality responses. In this work, we formalize this trade-off through the lens of regret minimization, demonstrating that the optimal strategy depends critically on the tail behavior of the reward distribution. We show theoretically that light-tailed regimes favor optimism to unearth high-quality outliers, whereas heavy-tailed regimes require pessimism to guard against reward mis-calibration in the extremes. Guided by this insight, we introduce Best-of-Tails (BoT), an adaptive inference-time alignment framework that uses Tsallis divergence as a tunable regularizer to provide a finer granularity of interpolation between these extremes. BoT uses the Hill estimator to characterize reward-tail heaviness on a per-prompt basis and dynamically adjusts its selection rule to balance exploration gains against alignment error. Across math, multiple-choice reasoning, and human-preference evaluations, BoT improves alignment performance across a range of reference and reward model configurations relative to fixed-strategy baselines.

Related papers

• LambdaPO: A Lambda Style Policy Optimization for Reasoning Language Models [34.349722314481824]
  Group Relative Policy Optimization is prized for its efficacy in foregoing an explicit value-critic.<n>The method's reliance on a monolithic statistical baseline, such as the group mean, collapses the relational topology of the trajectory space into a single scalar.<n>We introduce a novel framework, Lambda Policy Optimization (LambdaPO), that addresses this information-theoretic bottleneck.
  arXiv  Detail & Related papers  (2026-05-19T06:10:24Z)
• Temper and Tilt Lead to SLOP: Reward Hacking Mitigation with Inference-Time Alignment [20.603753314256704]
  Inference-time alignment techniques offer a lightweight alternative or complement to costly reinforcement learning.<n>We extend these techniques by introducing reference-model temperature adjustment.<n>To reward hacking, we propose an algorithm for calibrating SLOP weight parameters.
  arXiv  Detail & Related papers  (2026-05-13T13:47:06Z)
• Breaking the Curse of Repulsion: Optimistic Distributionally Robust Policy Optimization for Off-Policy Generative Recommendation [8.112649652437705]
  We argue that the solution lies in rigorously identifying the latent high-quality distribution entangled within a noisy behavior policy.<n>We prove that hard filtering is the exact solution to this DRO objective, enabling DRPO to optimally recover high-quality behaviors while strictly discarding divergence-inducing noise.
  arXiv  Detail & Related papers  (2026-02-11T02:18:27Z)
• MAESTRO: Meta-learning Adaptive Estimation of Scalarization Trade-offs for Reward Optimization [56.074760766965085]
  Group-Relative Policy Optimization has emerged as an efficient paradigm for aligning Large Language Models (LLMs)<n>We propose MAESTRO, which treats reward scalarization as a dynamic latent policy, leveraging the model's terminal hidden states as a semantic bottleneck.<n>We formulate this as a contextual bandit problem within a bi-level optimization framework, where a lightweight Conductor network co-evolves with the policy by utilizing group-relative advantages as a meta-reward signal.
  arXiv  Detail & Related papers  (2026-01-12T05:02:48Z)
• From Noisy Traces to Stable Gradients: Bias-Variance Optimized Preference Optimization for Aligning Large Reasoning Models [90.45197506653341]
  Large reasoning models generate intermediate reasoning traces before producing final answers.<n> aligning LRMs with human preferences, a crucial prerequisite for model deployment, remains underexplored.<n>A common workaround optimized a single sampled trajectory, which introduces substantial gradient variance from trace sampling.
  arXiv  Detail & Related papers  (2025-10-06T17:58:01Z)
• On Symmetric Losses for Robust Policy Optimization with Noisy Preferences [55.8615920580824]
  This work focuses on reward modeling, a core component in reinforcement learning from human feedback.<n>We propose a principled framework for robust policy optimization under noisy preferences.<n>We prove that symmetric losses enable successful policy optimization even under noisy labels.
  arXiv  Detail & Related papers  (2025-05-30T15:30:43Z)
• Self-Boost via Optimal Retraining: An Analysis via Approximate Message Passing [58.52119063742121]
  Retraining a model using its own predictions together with the original, potentially noisy labels is a well-known strategy for improving the model performance.<n>This paper addresses the question of how to optimally combine the model's predictions and the provided labels.<n>Our main contribution is the derivation of the Bayes optimal aggregator function to combine the current model's predictions and the given labels.
  arXiv  Detail & Related papers  (2025-05-21T07:16:44Z)
• Supervised Optimism Correction: Be Confident When LLMs Are Sure [91.7459076316849]
  We establish a novel theoretical connection between supervised fine-tuning and offline reinforcement learning.<n>We show that the widely used beam search method suffers from unacceptable over-optimism.<n>We propose Supervised Optimism Correction, which introduces a simple yet effective auxiliary loss for token-level $Q$-value estimations.
  arXiv  Detail & Related papers  (2025-04-10T07:50:03Z)
• Self-Evolutionary Large Language Models through Uncertainty-Enhanced Preference Optimization [9.618391485742968]
  Iterative preference optimization has recently become one of the de-facto training paradigms for large language models (LLMs) We present an uncertainty-enhanced textbfPreference textbfOptimization framework to make the LLM self-evolve with reliable feedback. Our framework substantially alleviates the noisy problem and improves the performance of iterative preference optimization.
  arXiv  Detail & Related papers  (2024-09-17T14:05:58Z)
• Provably Mitigating Overoptimization in RLHF: Your SFT Loss is Implicitly an Adversarial Regularizer [52.09480867526656]
  We identify the source of misalignment as a form of distributional shift and uncertainty in learning human preferences.<n>To mitigate overoptimization, we first propose a theoretical algorithm that chooses the best policy for an adversarially chosen reward model.<n>Using the equivalence between reward models and the corresponding optimal policy, the algorithm features a simple objective that combines a preference optimization loss and a supervised learning loss.
  arXiv  Detail & Related papers  (2024-05-26T05:38:50Z)
• Overcoming Reward Overoptimization via Adversarial Policy Optimization with Lightweight Uncertainty Estimation [46.61909578101735]
  Adversarial Policy Optimization (AdvPO) is a novel solution to the pervasive issue of reward over-optimization in Reinforcement Learning from Human Feedback. In this paper, we introduce a lightweight way to quantify uncertainties in rewards, relying solely on the last layer embeddings of the reward model.
  arXiv  Detail & Related papers  (2024-03-08T09:20:12Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.