arXiv cs.AI· Ziyi Ding, Xiaodi Huang, Xianxin Lai, Haijier Chen, Zhang Zhang, Xiao-Ping Zhang, Jiayu Chen·· 2 小时前AI 评分37
干预、校准、发现：从像素中学习因果结构
Intervene, Calibrate, Discover: Learning Causal Structure from Pixels
AI 导读
arXiv 论文提出一种从低层观测学习高层因果变量的方法：通过固定轮换干预获得经过验证的值，用可逆的维度级校准映射将图像读出与目标设置对齐到共享数值单位，再由智能体选择干预解析图方向并拟合局部机制。方法在 Domino、Pick-and-Place、CAUSAL3D、CausalWorld 等六个环境上评估图发现、干预后果与反事实图像；对照实验显示校准降低了同事实情境下其他变量预测变化的平均误差。
正文 · 原文
Abstract:Learning high-level causal variables from low-level observations is a central problem for artificial intelligence. With known node semantics but an unknown graph, visual readout errors can misalign model settings with environment intervention values. We address this mismatch by pairing image readouts with verified values attained through fixed round-robin interventions. With encoder and decoder frozen, a dimension-wise invertible calibration map puts visual responses and target settings in shared numerical units. After freezing this map, the agent selects interventions to resolve graph directions and fits local mechanisms to predict consequences; inverse calibration returns counterfactual states to the decoder's original coordinates. Across Domino, Pick-and-Place, Balance, Gears, CAUSAL3D, and CausalWorld, we evaluate graph discovery, intervention consequences, and counterfactual images. Controlled domino experiments show that interventions resolve orientation ambiguity left by equal additional observations. In a matched-data domino control, calibration reduces the mean error of predicted changes in other variables within the same factual context, relative to an uncalibrated model.
| Subjects: | Artificial Intelligence (cs.AI); Applications (stat.AP) |
| Cite as: | arXiv:2610.34711 [cs.AI] |
| (or arXiv:2610.34711v1 [cs.AI] for this version) | |
| https://doi.org/10.48550/arXiv.2610.34711 arXiv-issued DOI via DataCite (pending registration) |
Submission history
From: Ziyi Ding [view email]
[v1]
Mon, 26 Oct 2026 02:01:46 UTC (723 KB)
来源：arXiv cs.AI · arxiv.org