Saved in:
| Main Authors: | , , |
|---|---|
| Format: | Preprint |
| Published: |
2025
|
| Subjects: | |
| Online Access: | https://arxiv.org/abs/2505.20340 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866917453503660032 |
|---|---|
| author | Zhang, Yukun Dong, Qi Li, Mengkang |
| author_facet | Zhang, Yukun Dong, Qi Li, Mengkang |
| contents | Understanding how latent representations evolve during generation is a central open problem in large language model interpretability. We introduce \textbf{Dynamical Manifold Evolution Theory} (DMET), a phenomenological framework that models LLM generation as a controlled dynamical system evolving along a trajectory on a low-dimensional semantic manifold. DMET formalizes the structural correspondence between Transformer components and a first-order ODE governed by a semantic potential $V$, and characterizes trajectory geometry through three falsifiable proxy metrics: state continuity $C$, attractor clustering quality $Q$, and topological persistence $P$, targeting local smoothness, meso-scale basin structure, and global topological organization, respectively. Across six model architectures, four task types, and 1,080 experimental runs, all three metrics consistently predict text quality outcomes -- log-perplexity, grammaticality, and cross-sentence coherence -- after controlling for decoding parameters, with associations surviving Benjamini--Hochberg correction. Ablation and sanity-check experiments confirm that the effects arise from genuine trajectory structure rather than static distributional artefacts. Furthermore, online monitoring of $C$ drives an adaptive decoding controller that reduces perplexity from 48.5 to 14.6 relative to a fixed-parameter baseline, demonstrating that latent dynamics characterization translates directly into actionable generation control. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2505_20340 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | Latent Trajectory Dynamics in Large Language Models: A Manifold Evolution Framework with Empirical Validation Zhang, Yukun Dong, Qi Li, Mengkang Computation and Language Artificial Intelligence Understanding how latent representations evolve during generation is a central open problem in large language model interpretability. We introduce \textbf{Dynamical Manifold Evolution Theory} (DMET), a phenomenological framework that models LLM generation as a controlled dynamical system evolving along a trajectory on a low-dimensional semantic manifold. DMET formalizes the structural correspondence between Transformer components and a first-order ODE governed by a semantic potential $V$, and characterizes trajectory geometry through three falsifiable proxy metrics: state continuity $C$, attractor clustering quality $Q$, and topological persistence $P$, targeting local smoothness, meso-scale basin structure, and global topological organization, respectively. Across six model architectures, four task types, and 1,080 experimental runs, all three metrics consistently predict text quality outcomes -- log-perplexity, grammaticality, and cross-sentence coherence -- after controlling for decoding parameters, with associations surviving Benjamini--Hochberg correction. Ablation and sanity-check experiments confirm that the effects arise from genuine trajectory structure rather than static distributional artefacts. Furthermore, online monitoring of $C$ drives an adaptive decoding controller that reduces perplexity from 48.5 to 14.6 relative to a fixed-parameter baseline, demonstrating that latent dynamics characterization translates directly into actionable generation control. |
| title | Latent Trajectory Dynamics in Large Language Models: A Manifold Evolution Framework with Empirical Validation |
| topic | Computation and Language Artificial Intelligence |
| url | https://arxiv.org/abs/2505.20340 |