arXiv subjects

Boris Vidolov

Publications and source records attributed to Boris Vidolov.

5 recordsLinked to original sources

arXiv2026

Off-Context GRPO: Learning to Reason on Hard Problems using Privileged Information

Reinforcement learning with verifiable rewards (RLVR) improves reasoning in large language models. Yet, typical RLVR approaches fail on difficult problems: when a model cannot generate any correct solutions, it receives \textit{zero} learning signal. Providing privileged guidance during training, such as solution prefixes, can help overcome this learning cliff by steering the model towards {correct solutions with non-zero reward}. {We call these rollouts \textit{off-context}: they are generated from a training prompt that contains privileged guidance, while the target objective is defined by the original prompt without that guidance.} {We introduce} Off-Context GRPO (OC-GRPO), a minimally modified variant of GRPO that uses guided rollouts but applies an importance-corrected objective to steer the update back toward the original unguided objective, avoiding the mismatch that destabilizes uncorrected guided training. Empirically, our algorithm achieves a 3.8\% absolute improvement (13.7\% relative gain) over vanilla GRPO on average across standard mathematical reasoning benchmarks with negligible additional cost.

cs.LG↗

arXiv2026

Structure Enables Effective Self-Localization of Errors in LLMs

Self-correction in language models remains elusive. In this work, we explore whether language models can explicitly localize errors in incorrect reasoning, as a path toward building AI systems that can effectively correct themselves. We introduce a prompting method that structures reasoning as discrete, semantically coherent thought steps, and show that models can localize errors more reliably within this structure than in conventional, unstructured chain-of-thought reasoning. Motivated by how the human brain monitors errors at discrete decision points and resamples alternatives, we introduce Iterative Correction Sampling of Thoughts (Thought-ICS), a self-correction framework. Thought-ICS iteratively prompts the model to generate reasoning one discrete and complete thought at a time--where each thought represents a deliberate decision by the model--creating natural boundaries for precise error localization. Upon verification, the model localizes the first erroneous step, and the system backtracks to generate alternative reasoning from the last correct point. When asked to correct reasoning verified as incorrect by an oracle, Thought-ICS achieves 20-40% self-correction lift. In a completely autonomous setting without external verification, it outperforms contemporary self-correction baselines.

cs.AI↗

arXiv2026

Self-Improvement of Language Models by Post-Training on Multi-Agent Debate

Self-improvement, where models improve beyond their current performance without external supervision, remains a challenge. The core difficulty is sourcing a training signal stronger than what the model itself can currently produce. Majority voting has been shown to provide such a signal by aggregating over multiple samples, helping mitigate some of the inconsistencies in LM reasoning. In this work, we show that multi-agent debate--where models collaborate and exchange reasoning over multiple rounds--provides an even richer signal than single-round majority voting. We introduce Multi-Agent Consensus Alignment (MACA), which uses reinforcement learning (RL) to post-train models to effectively utilize multi-agent debate. We find that preference learning over full reasoning traces, learning to differentiate between majority and minority reasoning, is more effective than binary consensus rewards or SFT-based approaches for leveraging these debate signals. This produces three key improvements: models are (1) better at utilizing the multi-agent debate setting (+26.87% on MATH), (2) individually more accurate (+21.51% on MathQA), and (3) more self-consistent (+27.6% on GSM8K). We also see strong generalization to unseen benchmarks (+16.3% on GPQA, +11.6% on CommonsenseQA).

cs.AI↗

arXiv2025

Imbalanced Gradients in RL Post-Training of Multi-Task LLMs

Multi-task post-training of large language models (LLMs) is typically performed by mixing datasets from different tasks and optimizing them jointly. This approach implicitly assumes that all tasks contribute gradients of similar magnitudes; when this assumption fails, optimization becomes biased toward large-gradient tasks. In this paper, however, we show that this assumption fails in RL post-training: certain tasks produce significantly larger gradients, thus biasing updates toward those tasks. Such gradient imbalance would be justified only if larger gradients implied larger learning gains on the tasks (i.e., larger performance improvements) -- but we find this is not true. Large-gradient tasks can achieve similar or even much lower learning gains than small-gradient ones. Further analyses reveal that these gradient imbalances cannot be explained by typical training statistics such as training rewards or advantages, suggesting that they arise from the inherent differences between tasks. This cautions against naive dataset mixing and calls for future work on principled gradient-level corrections for LLMs.

cs.LG↗

arXiv2021

Decentralized Multi-Agent Pursuit using Deep Reinforcement Learning

Pursuit-evasion is the problem of capturing mobile targets with one or more pursuers. We use deep reinforcement learning for pursuing an omni-directional target with multiple, homogeneous agents that are subject to unicycle kinematic constraints. We use shared experience to train a policy for a given number of pursuers that is executed independently by each agent at run-time. The training benefits from curriculum learning, a sweeping-angle ordering to locally represent neighboring agents and encouraging good formations with reward structure that combines individual and group rewards. Simulated experiments with a reactive evader and up to eight pursuers show that our learning-based approach, with non-holonomic agents, performs on par with classical algorithms with omni-directional agents, and outperforms their non-holonomic adaptations. The learned policy is successfully transferred to the real world in a proof-of-concept demonstration with three motion-constrained pursuer drones.

cs.MA↗