Seek in the Dark: Reasoning via Test-Time Instance-Level Policy Gradient in Latent Space

Hengli Li, Chenxi Li, Tong Wu, Xuekai Zhu, Yuxuan Wang, Zhaoxin Yu, Eric Hanchen Jiang, Song-Chun Zhu, Zixia Jia, Ying Nian Wu^#, and Zilong Zheng^#

2025 · arXiv: arxiv.org/abs/2505.13308

Code Website

Abstract

Reasoning ability, a core component of human intelligence, continues to pose a significant challenge for Large Language Models (LLMs) in the pursuit of AGI. Although model performance has improved under the training scaling law, significant challenges remain, particularly with respect to training algorithms, such as catastrophic forgetting, and the limited availability of novel training data. As an alternative, test-time scaling enhances reasoning performance by increasing test-time computation without parameter updating. Unlike prior methods in this paradigm focused on token space, we propose leveraging latent space for more effective reasoning and better adherence to the test-time scaling law. We introduce LatentSeek, a novel framework that enhances LLM reasoning through Test-Time Instance-level Adaptation (TTIA) within the model’s latent space. Specifically, LatentSeek leverages policy gradient to iteratively update latent representations, guided by self-generated reward signals. LatentSeek is evaluated on a range of reasoning benchmarks, including GSM8K, MATH-500, and AIME2024, across multiple LLM architectures. Results show that LatentSeek consistently outperforms strong baselines, such as Chain-of-Thought prompting and fine-tuning-based methods. Furthermore, our analysis demonstrates that LatentSeek is highly efficient, typically converging within a few iterations for problems of average complexity, while also benefiting from additional iterations, thereby highlighting the potential of test-time scaling in the latent space. These findings position LatentSeek as a lightweight, scalable, and effective solution for enhancing the reasoning capabilities of LLMs.

Figure 1. Comparison of LatentSeek with RL-based fine-tuning and Prompt Engineering. RL-based fine-tuning methods generally require iterative updates to model parameters guided by reward signals. Prompt engineering approaches depend heavily on manually designed prompts. In contrast, LatentSeek performs optimization within the latent space.

Citation

@misc{li2025seekdarkreasoningtesttime,
      title={Seek in the Dark: Reasoning via Test-Time Instance-Level Policy Gradient in Latent Space},
      author={Hengli Li and Chenxi Li and Tong Wu and Xuekai Zhu and Yuxuan Wang and Zhaoxin Yu and Eric Hanchen Jiang and Song-Chun Zhu and Zixia Jia and Ying Nian Wu and Zilong Zheng},
      year={2025},
      eprint={2505.13308},
      archivePrefix={arXiv},
      primaryClass={cs.LG},
      url={https://arxiv.org/abs/2505.13308},
}

Related Publications

Zhao et al., Absolute Zero: Reinforced Self-play Reasoning with Zero Data, arXiv, 2025.

Zhang et al., Amulet: ReAlignment During Test Time for Personalized Preference Adaptation of LLMs, in ICLR, 2025.