arXivDaily arXiv每日学术速递 周一至周五更新
arXiv周末暂无论文更新,休息一下吧,周末愉快~~
arXiv 2602.08520cs.AIcs.LG
收藏

强化推断:利用不确定性进行自我修正的语言模型推理

Reinforcement Inference: Leveraging Uncertainty for Self-Correcting Language Model Reasoning

Programme of Management Engineering, Politecnico di Milano(米兰理工学院管理工程项目)

Xinhai Sun

AI总结:

本研究提出强化推断方法,通过利用模型不确定性进行自我修正推理,提升语言模型在零样本设置下的准确性,同时减少计算成本。

摘要:

现代大型语言模型(LLMs)通常在单次、贪婪推断协议下进行评估和部署,尤其是在需要确定性行为的专业领域。这种模式会系统性地低估固定模型的真实能力:许多错误并非源于知识缺失,而是源于内部模糊性下的过早承诺。我们引入了强化推断,这是一种熵感知的推断时间控制策略,利用模型自身的不确定性,有选择地调用第二次更谨慎的推理尝试,从而在不重新训练的情况下实现更强的性能。在12,032个MMLU-Pro问题上,使用DeepSeek-v3.2在零样本设置中进行确定性解码,强化推断将准确性从60.72%提高到84.03%,同时仅导致61.06%的额外推断调用。100%重新提问的消融实验达到84.35%,表明具有不确定性的选择捕获了大部分可实现的改进,同时计算成本显著降低。此外,仅使用提示的消融实验表现不佳,表明收益并非由通用提示单独解释。除了提供一种实用的推断时间升级外,我们的结果还表明了一种更广泛的熵感知范式,用于衡量和扩展模型能力:因为现代基于解码器的模型生成输出是自回归的,熵和相关置信度度量自然成为生成过程中的首要控制信号。由此产生的单次贪婪推断与不确定性条件下的深入推理之间的差距为LLM的潜在推理范围提供了诊断视角,并激励了未来明确约束正确性-置信度一致性的训练目标。

英文摘要:

Modern large language models (LLMs) are often evaluated and deployed under a one-shot, greedy inference protocol, especially in professional settings that require deterministic behavior. This regime can systematically under-estimate a fixed model's true capability: many errors arise not from missing knowledge, but from premature commitment under internal ambiguity. We introduce Reinforcement Inference, an entropy-aware inference-time control strategy that uses the model's own uncertainty to selectively invoke a second, more deliberate reasoning attempt, enabling stronger performance without any retraining. On 12,032 MMLU-Pro questions across 14 subjects, using DeepSeek-v3.2 with deterministic decoding in a zero-shot setting, Reinforcement Inference improves accuracy from 60.72% to 84.03%, while only incurring 61.06% additional inference calls. A 100% re-asking ablation reaches 84.35%, indicating that uncertainty-aware selection captures most of the attainable improvement with substantially less compute. Moreover, a prompt-only ablation underperforms the baseline, suggesting that the gains are not explained by generic prompting alone. Beyond providing a practical inference-time upgrade, our results suggest a broader entropy-aware paradigm for measuring and expanding model capability: because modern decoder-based models generate outputs autoregressively, entropy and related confidence measures arise naturally as first-class control signals during generation. The resulting gap between one-pass greedy inference and uncertainty-conditioned deliberation offers a diagnostic lens on an LLM's latent reasoning horizon and motivates future training objectives that explicitly constrain correctness--confidence alignment.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑