Abstract:Recent work on recursive architectures has shown that tiny neural networks can be surprisingly powerful on structured reasoning tasks. The trick is to model reasoning trajectories with a latent dynamical system. We argue that the inference-time behaviour of these architectures is best understood as approximate inference over latent reasoning trajectories, with deterministic recursion as the one-particle, zero-noise limit. We make this view operational through guided stochastic exploration: stochastic perturbations of the reasoning dynamics propose neighbouring trajectories, and the model's existing early-stopping head reweights them online. The framework yields three label-free diagnostics: local stability, guide alignment, and cloud-token entropy. These predict, from inference traces alone, whether the procedure will help and which of its outputs to trust. On Sudoku-Extreme it lifts exact-solve accuracy from $85.9\%$ to $98.0\%$ without retraining; on Maze-Hard the diagnostics flag a misaligned guide, as validation performance later confirms. The same machinery thus characterises both when recursive reasoning has room to improve at the trajectory level and when the model's internal guide can recover it.
| Comments: | Presented at the proceedings of the ICML 2026 Workshop on Structured Probabilistic Inference & Generative Modeling (SPIGM)}, Seoul, South Korea. 2026 |
| Subjects: | Artificial Intelligence (cs.AI) |
| Cite as: | arXiv:2605.25230 [cs.AI] |
| (or arXiv:2605.25230v1 [cs.AI] for this version) | |
| https://doi.org/10.48550/arXiv.2605.25230 arXiv-issued DOI via DataCite (pending registration) |
From: Andrew Corbett Dr [view email]
[v1]
Sun, 24 May 2026 19:32:20 UTC (315 KB)
此内容由惯性聚合(RSS阅读器)自动聚合整理，仅供阅读参考。 原文来自 — 版权归原作者所有。