Author

Jian Liang

2 papers indexed here

We haven’t gathered this author’s papers yet. Follow them and we’ll fetch their work.

Not the right person? Other researchers publish under this name.

#artificial intelligence Jan 2026

Do Latent-CoT Models Think Step-by-Step? A Mechanistic Study on Sequential Reasoning Tasks

It is shown theoretically that the task's algebraic structure controls its effective memory: compressible regimes support late-bottleneck reasoning, while incompressible regimes preserve full-history dependence and destabilize latent rollouts.

Jian Liang, Liang-Ming Pan · arXiv.org · 12 citations

#artificial intelligence Preprint Sep 2026

Beyond the Training Horizon: Mechanisms and Limits of Length Generalization in Looped Transformers

It is shown that length generalization need not rely on faithful step-by-step reasoning, as Looped Transformers can exploit task structure without explicitly representing every intermediate state.

Jian Liang, Xi Jin, Liang-Ming Pan · 0 citations