论文研究
x.com ·
@omarsar0
Looped Flows：用局部去噪目标训练循环推理模型
一篇论文提出 Looped Flows 训练方法，用于改进循环（looped）模型的推理能力。循环模型在推理时反复更新隐藏状态，可在不增加参数的情况下获得更多推理计算，但训练困难：梯度通常只流经最后一到两次更新，早期更新无法学会为后续更新做准备。该方法借鉴流模型的训练方式，用局部去噪目标训练循环过程，噪声水平逐步降低并共享同一噪声样本，从而把每次更新与下一次关联起来。推理时模型沿概率流进行，更细的时间网格会消耗更多计算，不同初始噪声可在多解任务上给出不同有效答案。作者报告在六个推理基准中的五个上优于此前的循环模型。
AI推荐理由 该帖介绍了一篇关于改进循环推理的论文（arXiv:2609.11801），核心新信息是提出 Looped Flows 训练方法：用局部去噪目标训练循环结构，使噪声水平逐步下降并共享同一噪声样本，从而把每次更新与下一次绑定，缓解梯度只流经最后一步、早期更新学不到东西的问题；推理时沿概率流进行，更细的时间网格意味着更多计算，不同初始噪声可在多解任务上产生不同有效答案。作者报告在六个推理基准中的五个上优于既有循环模型。局限在于：这些结论均来自论文作者自述，帖子未提供独立复现或评测细节，且互动量很低（930 浏览、5 赞），仅能作为传播信号，不能替代质量判断。对关注推理架构、训练方法和算力-推理权衡的研究者与开发者有参考价值。
72
浏览 930
点赞 5
评论 4
转发 0
正文
Interesting paper to improve recurrent reasoning.
Looped models are great because you get more reasoning out of a model without adding parameters.
So this work proposes a looped architecture with a new training method.
The authors report wins over prior looped models on five of six reasoning benchmarks.
More details from the paper:
Looped models reason by updating a hidden state again and again at inference time. The hard part is training. Gradients usually flow through only the last one or two updates, so the early updates never learn to set up the later ones.
Looped flows train the recurrence with local denoising objectives, the way flow models are trained. Noise levels decrease step by step and share the same noise sample, which ties each update to the next.
At inference the model follows a probability flow. A finer time grid spends more compute, and different starting noise can produce different valid answers on tasks with more than one solution.
Paper: [arxiv.org/abs/2609.11801](https://arxiv.org/abs/2609.11801)
Chat with Paper: [academy.dair.ai/papers/thinkin…](https://academy.dair.ai/papers/thinking-with-looped-flows-2609.11801)
