arXiv cs.CL· Siheng Xiong, Ali Payani, Faramarz Fekri·· 15 小时前AI 评分34
多路径规划聚合（MPPA）推进测试时扩展的帕累托前沿
Advancing the Pareto Frontier of Test-Time Scaling with Multi-Path Plan Aggregation
AI 导读
多路径规划聚合（MPPA）在共享冻结主干上分别用 LoRA 适配器学习规划与答案选择，规划器联合生成 K 个简短计划，各自引导冻结执行器生成解答，再由聚合器打分选出最终答案。在五个数学与科学基准、三个主干模型上，MPPA 用 3 个计划即超过 8 个直接采样加多数投票或同等学习聚合器的平均准确率，输出 token 减少 47–54%。
正文 · 原文
Abstract:Test-time scaling seeks to improve reasoning through additional inference-time computation, commonly using solution sampling with majority voting to select the final answer. However, repeated sampling can reproduce correlated errors, limiting the accuracy gains from additional computation. We introduce Multi-Path Plan Aggregation (MPPA), which learns planning and answer selection using separate LoRA adapters on a shared frozen backbone. The planner jointly generates $K$ short plans, each guiding a separate solution generated by the frozen executor; the aggregator then scores the resulting solutions and selects the final answer. Joint planning encourages complementary approaches suited to the executor's capabilities, while learned selection uses reasoning evidence to identify correct candidates. Alternating optimization improves planning and aggregation through supervised finetuning and reinforcement learning. Across five mathematics and science benchmarks and three backbones, MPPA improves accuracy--token trade-offs compared with training-free and training-based baselines. With three plans, it achieves higher average accuracy than eight directly sampled solutions with majority voting or the same learned aggregator, while reducing output tokens by 47--54%. Evaluation with task-provided subtasks further demonstrates higher final-task accuracy than direct sampling with majority voting or learned aggregation at either the task or subtask level.
| Subjects: | Computation and Language (cs.CL) |
| Cite as: | arXiv:2510.11620 [cs.CL] |
| (or arXiv:2510.11620v3 [cs.CL] for this version) | |
| https://doi.org/10.48550/arXiv.2510.11620 arXiv-issued DOI via DataCite |
Submission history
From: Siheng Xiong [view email]
[v1]
Mon, 13 Oct 2025 17:02:41 UTC (269 KB)
[v2]
Thu, 16 Oct 2025 21:41:15 UTC (270 KB)
[v3]
Mon, 26 Oct 2026 07:08:48 UTC (211 KB)
来源：arXiv cs.CL · arxiv.org