Sparse autoencoders (SAEs) improve interpretability in multimodal models, but it remains unclear whether SAE features form modular, composable units for reasoning-an assumption underlying many intervention-based steering methods. We test this modularity hypothesis and find it often fails: intervening on a task-selective feature set can modestly improve reasoning accuracy, while intervening on the union of two such sets reliably induces output drift (large unintended changes in predictions) and degrades accuracy, even under norm-matched perturbations. This non modular circuit interference is consistent with shared internal pathways where feature unions amplify activation shifts. We develop a reproducible causal pipeline to localize and test these sparse visual thought circuits in Qwen3-VL-8B. On a controlled synthetic benchmark with seven task types and three difficulty levels, linear probes identify a mid decoder locus for task type information. We train SAEs at this layer, construct task-selective sets via an explicit rule, and perform inference time scaling and ablation while quantifying accuracy and drift. Our findings-validated with bootstrapped subsamples and permutation controls, and replicated across multiple VLM families and five diverse datasets clarify the boundaries of SAE feature composability and provide a rigorous diagnostic framework for more reliable VLM control.
Detecting Unfaithful Chain-of-Thought via Circuit-Guided Internal-External Discrepancy
Jilin University（吉林大学）；University of Central Florida（中央佛罗里达大学）；Arizona State University（亚利桑那州立大学）；University of Vienna（维也纳大学）；University of North Carolina at Chapel Hill（北卡罗来纳大学教堂山分校）
SCAN: Sparse Circuit Anchor Interpretable Neuron for Lifelong Knowledge Editing
New Laboratory of Pattern Recognition (NLPR), State Key Laboratory of Multimodal Artificial Intelligence Systems (MAIS), Institute of Automation, Chinese Academy of Sciences（模式识别新实验室（NLPR）、多模态人工智能系统国家重点实验室（MAIS）、自动化研究所、中国科学院）；Cuiying Honors College, Lanzhou University（cuiying荣誉学院、兰州大学）；Research Institute of Intelligent Complex Systems, Fudan University（智能复杂系统研究院、复旦大学）；Zhongguancun Academy（中关村学院）
TCPO: Thought-Centric Preference Optimization for Effective Embodied Decision-making
Tsinghua University（清华大学）；Meituan（美团）；Northeastern University（东北大学）；Beijing University of Posts and Telecommunications（北京邮电大学）；Meta AI；Wuhan University（武汉大学）
Symbolic Analysis of Grover Search Algorithm via Chain-of-Thought Reasoning and Quantum-Native Tokenization
Department of Computer Science, The University of Pittsburgh（匹兹堡大学计算机科学系）；Department of Computer Science, The University of North Carolina at Chapel Hill（北卡罗来纳大学教堂山分校计算机科学系）
COTCAgent: Preventive Consultation via Probabilistic Chain-of-Thought Completion
School of Computing and Data Science, The University of Hong Kong（香港大学计算与数据科学学院）；Shenzhen Institute for Advanced Study, University of Electronic Science and Technology of China（中国电子科学与技术大学深圳高级研究所）；School of Computer Science, The University of Sydney（悉尼大学计算机科学学院）
Sparse Probabilistic Circuits via Pruning and Growing
UCLA（加州大学洛杉矶分校）
概率电路（PC）是一种易于处理的概率分布表示，允许精确和有效地计算似然和边缘。最近在提高PC的规模和表现力方面取得了重大进展。然而，PC训练性能随着模型大小的增加而趋于稳定。我们发现，在现有的大型PC结构的大部分容量是浪费：全连接的参数层，只有稀疏使用。我们提出了两个操作：修剪和增长，利用PC结构的稀疏性。具体地，修剪操作移除PC的不重要的子网络用于模型压缩，并且具有理论保证。增长操作通过增加潜在空间的大小来增加模型容量。通过交替应用修剪和增长，我们增加了有意义地使用的容量，使我们能够显着扩展PC学习。从经验上讲，与其他PC学习器和不太容易处理的深度生成模型（如基于流的模型和变分自编码器（VAE））相比，我们的学习器在MNIST系列图像数据集和Penn Tree Bank语言数据上实现了最先进的可能性。
SCOUT: Self-Checking and Recovery-Aware Tool-Thought Agents for Ultra-Long Egocentric Video Reasoning
Sun Yat-sen University（中山大学）；Shenzhen Loop Area Institute（深圳河套学院）；Guangdong OPPO Mobile Telecommunications Corp., Ltd.（广东欧珀移动通信有限公司）；OPPO AI Center（OPPO人工智能中心）；The Chinese University of Hong Kong, Shenzhen（香港中文大学（深圳））
Neuronal Attention Circuit (NAC) for Representation Learning
Department of Automation, University of Science \& Technology of China, Hefei, China；Institute of Artificial Intelligence, Hefei Comprehensive National Science Center
