arXiv Thu batch; unlabeled VLM training · P2 · 2026-07-11

BUS：它和通用视觉自监督的关系在于：通过无标签 backward prediction 增强 VLM reasoning，属于视觉-语言自监督式训练信号

中相关；详见方法、贡献和实验边界。

BUS: Brain-Inspired Unsupervised Self-Reflection for Advanced Multimodal Reasoning　arXiv Thu batch; unlabeled VLM training　原文链接

编号2607.07361 优先级P2 类别arXiv Thu batch; unlabeled VLM training 会议arXiv Thu batch; unlabeled VLM training 方法通过无标签 backward prediction 增强 VLM reasoning，属于视觉-语言自监督式训练信号来源arXiv / OpenReview

先说结论。它和通用视觉自监督的关系在于：通过无标签 backward prediction 增强 VLM reasoning，属于视觉-语言自监督式训练信号。中相关；详见方法、贡献和实验边界。

**(c) Evidence of Backward Prediction Figure 2: VLMs perform backward prediction**(c) Evidence of Backward Prediction Figure 2: VLMs perform backward prediction. (a) Task environment. States are represented by images, and darker arrows denote higher state-to-state transition probabilities. The experiment begins with a learning phase in which all models are presented with state transitions. (b) Decision phase. Rewards are placed in three states to dissociate backward prediction and forward prediction. (c) Evidence of backward prediction. Among these models, at least 65% of the choices are consistent with backward prediction.这张图来自论文 PDF 的结构化抽取。当前用于辅助理解 BUS 的方法或实验，请结合正文精读段落一起看。

**Figureure 1 · : Backward prediction process in question-answering scenarios**Figure 1: Backward prediction process in question-answering scenarios. To answer the given question, the brain predicts which events are likely to precede an image that possibly contains a cat.这张图来自论文 PDF 的结构化抽取。当前用于辅助理解 BUS 的方法或实验，请结合正文精读段落一起看。

核心问题

它和通用视觉自监督的关系在于：通过无标签 backward prediction 增强 VLM reasoning，属于视觉-语言自监督式训练信号。

方法拆解

通过无标签 backward prediction 增强 VLM reasoning，属于视觉-语言自监督式训练信号

主要贡献

中相关；详见方法、贡献和实验边界。

实验看点

实验部分建议重点看两类证据：一是作者是否把方法收益和更强数据、更长训练、更大模型区分开；二是跨模型、跨数据或跨任务迁移是否还能保留同样趋势。

局限与读法

这篇论文的结论需要结合任务设置、训练数据规模和消融实验一起看；不要只凭单个指标判断它对通用视觉表征的价值。