arXiv cs.AI· Junyeob Baek, Mingyu Jo, Minsu Kim, Mengye Ren, Yoshua Bengio, Sungjin Ahn·· 4 小时前AI 评分39
生成式递归推理模型 GRAM：将递归潜在推理转化为概率化多轨迹计算
Generative Recursive Reasoning
AI 导读
arXiv 论文提出生成式递归推理模型 GRAM，把递归潜在推理变成随机的多轨迹概率计算，支持多假设、多解策略，并可通过递归深度与并行轨迹采样实现推理时扩展。GRAM 用摊销变分推理训练，在结构化推理和多解约束满足任务上优于确定性递归基线，还能进行无条件生成。
正文 · 原文
Abstract:How should future neural reasoning systems implement extended computation? Recursive Reasoning Models (RRMs) offer a promising alternative to autoregressive sequence extension by performing iterative latent-state refinement with shared transition functions. Yet existing RRMs are largely deterministic, following a single latent trajectory and converging to a single prediction. We introduce Generative Recursive reAsoning Models (GRAM), a framework that turns recursive latent reasoning into probabilistic multi-trajectory computation. GRAM models reasoning as a stochastic latent trajectory, enabling multiple hypotheses, alternative solution strategies, and inference-time scaling through both recursive depth and parallel trajectory sampling. This yields a latent-variable generative model supporting conditional reasoning via $p_\theta(y \mid x)$ and, with fixed or absent inputs, unconditional generation via $p_\theta(x)$. Trained with amortized variational inference, GRAM improves over deterministic recurrent and recursive baselines on structured reasoning and multi-solution constraint satisfaction tasks, while demonstrating an unconditional generation capability. this https URL
| Subjects: | Artificial Intelligence (cs.AI) |
| Cite as: | arXiv:2605.19376 [cs.AI] |
| (or arXiv:2605.19376v5 [cs.AI] for this version) | |
| https://doi.org/10.48550/arXiv.2605.19376 arXiv-issued DOI via DataCite |
Submission history
From: Junyeob Baek [view email]
[v1]
Tue, 19 May 2026 05:20:56 UTC (7,933 KB)
[v2]
Wed, 20 May 2026 08:03:28 UTC (7,933 KB)
[v3]
Wed, 7 Oct 2026 14:00:19 UTC (4,673 KB)
[v4]
Thu, 15 Oct 2026 15:32:05 UTC (4,673 KB)
[v5]
Fri, 30 Oct 2026 02:20:39 UTC (4,673 KB)
来源：arXiv cs.AI · arxiv.org