When to Re-Plan: Subgoal Persistence in Hierarchical Latent Reasoning
The paper titled 'When to Re-Plan: Subgoal Persistence in Hierarchical Latent Reasoning' explores the balance between stability and adaptability in long-horizon reasoning systems. It introduces a feudal-style manager-worker interface to enhance subgoal persistence, which is crucial for effective multi-step computation. The findings suggest that moderate periods of subgoal persistence lead to better performance compared to very frequent or very long planning horizons.
- ▪The study focuses on the stability-adaptivity tradeoff in latent reasoning systems.
- ▪A feudal-style manager-worker interface is proposed to improve subgoal persistence.
- ▪Moderate periods of subgoal persistence consistently outperform both very frequent and very long planning horizons.
arXiv cs.AI files mainly under ai research. We currently carry 1,128 of its stories.
Opening excerpt (first ~120 words) tap to expand
Computer Science > Artificial Intelligence arXiv:2606.03741 (cs) [Submitted on 2 Jun 2026] Title:When to Re-Plan: Subgoal Persistence in Hierarchical Latent Reasoning Authors:Ayushi Chadha View a PDF of the paper titled When to Re-Plan: Subgoal Persistence in Hierarchical Latent Reasoning, by Ayushi Chadha View PDF HTML (experimental) Abstract:Long-horizon reasoning requires a system to commit to medium-horizon intent without becoming rigid: re-plan too often and computation never coheres into multi-step structure; commit too long and the plan goes stale. We study this stability-adaptivity tradeoff in the latent reasoning setting, where multi-step computation occurs inside hidden state rather than externalized token traces.
…
Excerpt limited to ~120 words for fair-use compliance. The full article is at arXiv cs.AI.
Discussion
0 commentsMore from arXiv cs.AI
