Architectural Proprioception in State Space Models: Thermodynamic Training Induces Anticipatory Halt Detection
作者: Jay Noon
分类: cs.LG, cs.AI
发布日期: 2026-03-04
备注: 17 pages, 15 figures
💡 一句话要点
提出概率导航架构以提升状态空间模型的自我意识
🎯 匹配领域: 支柱二:RL算法与架构 (RL & Architecture) 支柱九:具身大模型 (Embodied Foundation Models)
关键词: 状态空间模型 热力学训练 概率导航架构 自我意识 计算效率 元认知 停止信号检测
📋 核心要点
- 现有的状态空间模型在处理复杂任务时缺乏自我意识,导致计算效率低下。
- 本文提出了一种基于热力学原则的概率导航架构,通过热力学损失函数优化模型训练,提升自我感知能力。
- 实验结果显示,热力学训练的SSMs在停止信号检测上表现优异,F1分数达到94.5%,显著优于变换器的86.4%。
📝 摘要(中文)
本文介绍了一种概率导航架构(PNA)框架,将神经计算视为通过热力学原理支配的概率流形的导航。我们使用一种新颖的热力学损失函数训练状态空间模型(SSMs)和变换器,惩罚计算浪费。实验结果表明,经过热力学训练的SSMs展现出建筑本体感知,表现为递归状态熵与停止信号之间的强预期耦合。该现象在不同任务中具有普遍性,且与变换器的表现显著不同,表明这一现象依赖于模型架构。我们的研究为成本感知推理和动态令牌预算提供了新的视角。
🔬 方法详解
问题定义:本文旨在解决状态空间模型在复杂任务中缺乏自我意识和计算效率低下的问题。现有方法未能有效利用热力学原理来优化模型性能,导致计算资源浪费。
核心思路:论文提出的核心思路是通过热力学损失函数来训练模型,惩罚计算浪费,从而提升模型的自我感知能力。该设计使得模型能够在处理信息时更具前瞻性,优化决策过程。
技术框架:整体架构包括概率导航架构(PNA),通过热力学损失函数训练状态空间模型(SSMs)和变换器。主要模块包括状态熵计算、停止信号检测和热力学压力调节。
关键创新:最重要的技术创新点在于引入热力学损失函数,使得模型不仅关注准确性,还关注计算效率。这一方法与传统的交叉熵损失函数相比,提供了更全面的优化目标。
关键设计:关键参数设置包括能量惩罚(alpha)和停止监督(beta),通过2D超参数搜索优化这些参数,发现热力学压力是主要诱导机制,而显式停止监督则起到放大作用。
📊 实验亮点
实验结果显示,热力学训练的状态空间模型在停止信号检测上表现出色,F1分数为94.5%,而变换器为86.4%。此外,SSMs在跨任务转移实验中表现出更强的元认知能力,零样本转移F1为64.2%,而变换器为69.3%。
🎯 应用场景
该研究的潜在应用领域包括智能机器人、自动驾驶、自然语言处理等需要高效决策的系统。通过提升模型的自我意识和计算效率,可以在实际应用中实现更高的性能和更低的资源消耗,具有重要的实际价值和未来影响。
📄 摘要(原文)
We introduce the Probability Navigation Architecture (PNA) framework, which treats neural computation as navigation through a probability manifold governed by thermodynamic principles. We train State Space Models (SSMs) and Transformers with a novel thermodynamic loss function that penalizes computational waste alongside standard cross-entropy. Across 19 experimental phases, we discover that thermodynamically-trained SSMs develop architectural proprioception: a strong anticipatory coupling between recurrent state entropy and halt confidence (r = -0.836, p < 0.001) in which the halt signal leads state entropy collapse by exactly two tokens (tau = -2.0). This Universal Stopping Signature (USS) reproduces to four decimal places across random seeds and generalizes to a structurally distinct sorting task. Critically, Transformers trained identically show no such coupling (r = -0.07), demonstrating that the phenomenon is architecture-dependent. Cross-task transfer experiments confirm that SSM halt detection reflects genuine meta-cognition (zero-shot transfer F1: SSMs 64.2% vs. Transformers 69.3%; post-adaptation: SSMs 94.5% vs. Transformers 86.4%), while Transformer halt detection relies on syntactic pattern matching. A 2D hyperparameter sweep over energy penalty (alpha) and halt supervision (beta) reveals that the anticipatory coupling is continuously controllable through training, with thermodynamic pressure serving as the primary induction mechanism and explicit halt supervision as an amplifier. Our results establish that SSMs are thermodynamically native architectures whose fixed-size recurrent states naturally support the Markovian compression that enables computational self-awareness, with implications for cost-aware inference, dynamic token budgets, and confidence-based routing in production systems.