arXiv cs.AI· Shahar Haim, Daniel C. McNamee·· 4 小时前AI 评分31
预训练表征中的直觉向量如何引导扩散模型推理
Intuition Vectors from Pretrained Representations Steer Diffusion-Based Reasoning
AI 导读
研究者用冻结的自监督视觉编码器提取“直觉向量”,即图像嵌入间的简单向量差,可区分 794 个抽象推理任务,准确率最高 87.2%。在 ARC 任务上,用直觉向量条件化预训练视觉潜空间的扩散 Transformer,效果优于完整示例条件化,测试时算力减少近 6 倍,并支持零样本推理。
正文 · 原文
Abstract:Abstract visual reasoning requires identifying relationships between few examples and applying them to novel inputs. Recent advances have been driven by large language and vision-language models or specialized domain-specific architectures. It remains unclear whether abstract reasoning can be supported directly by the relational geometry of pretrained visual representations, without relying on language-based inference or task-specific representation learning. We investigate this possibility using frozen self-supervised vision encoders. With latent vector arithmetic, we confirmed that the representational geometry of two self-supervised encoders support abstract concept discrimination on Bongard problems and encode systematic relationships between input-output pairs in ARC tasks. Simple vector differences between image embeddings, which we call intuition vectors, capture transformations that generalize across instances of the same task. These vectors distinguish among 794 abstract reasoning tasks with up to 87.2\% accuracy, preserve relational structure across task-invariant visual transformations, and guide output retrieval for unseen inputs. Given this, we tested the hypothesis that this relational structure can steer generative reasoning by conditioning a diffusion transformer operating in pretrained visual latent space. On held-out ARC tasks, intuition-vector conditioning outperforms conditioning on complete demonstrations while requiring nearly six times less test-time compute, and supports zero-shot inference. Taken together, our findings suggest that pretrained visual representations provide a transferable substrate for abstract reasoning, enabling a modular approach in which the relational structure inferred from demonstrations steer generative reasoning.
| Subjects: | Machine Learning (cs.LG); Artificial Intelligence (cs.AI) |
| Cite as: | arXiv:2609.32619 [cs.LG] |
| (or arXiv:2609.32619v2 [cs.LG] for this version) | |
| https://doi.org/10.48550/arXiv.2609.32619 arXiv-issued DOI via DataCite |
Submission history
From: Shahar Haim [view email]
[v1]
Sat, 26 Sep 2026 13:40:07 UTC (10,351 KB)
[v2]
Tue, 17 Nov 2026 14:06:33 UTC (14,757 KB)
来源:arXiv cs.AI · arxiv.org