With this paper, we survey techniques for improving the predictive accuracy of pretrained large language models by allocating additional compute at inference time. In categorizing test-time scaling methods, we place special emphasis on how a problem is decomposed into subproblems and on the topological organization of these subproblems whether sequential, parallel, or tree-structured. This perspective allows us to unify diverse approaches such as Chain-of-Thought, Branch-Solve-Merge, and Tree-of-Thought under a common lens. We further synthesize existing analyses of these techniques, highlighting their respective strengths and weaknesses, and conclude by outlining promising directions for future research
@article{arxiv.2511.14772,
title = {Test-time Scaling of LLMs: A Survey from A Subproblem Structure Perspective},
author = {Zhuoyi Yang and Xu Guo and Tong Zhang and Huijuan Xu and Boyang Li},
journal= {arXiv preprint arXiv:2511.14772},
year = {2025}
}