Cover Image

Counting Circuits: Mechanistic Interpretability of Visual Reasoning in Large Vision-Language Models

Fuente: arXiv

Saved in:

Bibliographic Details
Main Authors:	Che, Liwei, Xue, Zhiyu, Quan, Yihao, Liu, Benlin, Shi, Zeru, Hurst, Michelle, Feldman, Jacob, Tang, Ruixiang, Krishna, Ranjay, Pavlovic, Vladimir
Format:	Preprint
Published:	2026
Subjects:	Computer Vision and Pattern Recognition Artificial Intelligence
Online Access:	Acceder al recurso
Tags:	Add Tag No Tags, Be the first to tag this record!

Similar Items

Improving Visual Reasoning with Iterative Evidence Refinement
by: Shi, Zeru, et al.
Published: (2026)

Reinforcing Consistency in Video MLLMs with Structured Rewards
by: Quan, Yihao, et al.
Published: (2026)

Hallucinatory Image Tokens: A Training-free EAZY Approach on Detecting and Mitigating Object Hallucinations in LVLMs
by: Che, Liwei, et al.
Published: (2025)

Read the Scene, Not the Script: Outcome-Aware Safety for LLMs
by: Wu, Rui, et al.
Published: (2025)

MemEye: A Visual-Centric Evaluation Framework for Multimodal Agent Memory
by: Guo, Minghao, et al.
Published: (2026)

Visual Representations inside the Language Model
by: Liu, Benlin, et al.
Published: (2025)

CapNav: Benchmarking Vision Language Models on Capability-conditioned Indoor Navigation
by: Su, Xia, et al.
Published: (2026)

Coarse Correspondences Boost Spatial-Temporal Reasoning in Multimodal Language Model
by: Liu, Benlin, et al.
Published: (2024)

Efficient Inference of Vision Instruction-Following Models with Elastic Cache
by: Liu, Zuyan, et al.
Published: (2024)

Where Fake Citations Are Made: Tracing Field-Level Hallucination to Specific Neurons in LLMs
by: Chen, Yuefei, et al.
Published: (2026)

Seeking and Updating with Live Visual Knowledge
by: Fu, Mingyang, et al.
Published: (2025)

Seeing Through Circuits: Faithful Mechanistic Interpretability for Vision Transformers
by: Żukowska, Nina, et al.
Published: (2026)

Meaningless Tokens, Meaningful Gains: How Activation Shifts Enhance LLM Reasoning
by: Shi, Zeru, et al.
Published: (2025)

A Single Layer to Explain Them All:Understanding Massive Activations in Large Language Models
by: Shi, Zeru, et al.
Published: (2026)

AgentForesight: Online Auditing for Early Failure Prediction in Multi-Agent Systems
by: Zhang, Boxuan, et al.
Published: (2026)

Structure From Tracking: Distilling Structure-Preserving Motion for Video Generation
by: Fei, Yang, et al.
Published: (2025)

Reasoning Circuits in Language Models: A Mechanistic Interpretation of Syllogistic Inference
by: Kim, Geonhee, et al.
Published: (2024)

Visual Program Distillation: Distilling Tools and Programmatic Reasoning into Vision-Language Models
by: Hu, Yushi, et al.
Published: (2023)

Ablate-to-Validate: Are Vision-Language Models Really Using Continuous Thought Tokens?
by: Zhang, Tianyi, et al.
Published: (2026)

AdaReasoner: Dynamic Tool Orchestration for Iterative Visual Reasoning
by: Song, Mingyang, et al.
Published: (2026)

Adaptive Circuit Behavior and Generalization in Mechanistic Interpretability
by: Nainani, Jatin, et al.
Published: (2024)

Efficient and Interpretable Transformer for Counterfactual Fairness
by: Dong, Panyi, et al.
Published: (2026)

PerceptionComp: A Video Benchmark for Complex Perception-Centric Reasoning
by: Li, Shaoxuan, et al.
Published: (2026)

Iterated Learning Improves Compositionality in Large Vision-Language Models
by: Zheng, Chenhao, et al.
Published: (2024)

Scale Can't Overcome Pragmatics: The Impact of Reporting Bias on Vision-Language Reasoning
by: Kamath, Amita, et al.
Published: (2026)

FailSafe: Reasoning and Recovery from Failures in Vision-Language-Action Models
by: Lin, Zijun, et al.
Published: (2025)

Beyond Transcription: Mechanistic Interpretability in ASR
by: Glazer, Neta, et al.
Published: (2025)

Formal Mechanistic Interpretability: Automated Circuit Discovery with Provable Guarantees
by: Hadad, Itamar, et al.
Published: (2026)

Perception Tokens Enhance Visual Reasoning in Multimodal Language Models
by: Bigverdi, Mahtab, et al.
Published: (2024)

Mechanistically Interpreting Compression in Vision-Language Models
by: Elluru, Veeraraju, et al.
Published: (2026)

MINAR: Mechanistic Interpretability for Neural Algorithmic Reasoning
by: He, Jesse, et al.
Published: (2026)

The Hard Positive Truth about Vision-Language Compositionality
by: Kamath, Amita, et al.
Published: (2024)

Mechanistic Interpretability of Diffusion Models: Circuit-Level Analysis and Causal Validation
by: Roy, Dip
Published: (2025)

SPARO: Selective Attention for Robust and Compositional Transformer Encodings for Vision
by: Vani, Ankit, et al.
Published: (2024)

RefDecoder: Enhancing Visual Generation with Conditional Video Decoding
by: Fan, Xiang, et al.
Published: (2026)

Dictionary Learning Improves Patch-Free Circuit Discovery in Mechanistic Interpretability: A Case Study on Othello-GPT
by: He, Zhengfu, et al.
Published: (2024)

Evaluating Brain-Inspired Modular Training in Automated Circuit Discovery for Mechanistic Interpretability
by: Nainani, Jatin
Published: (2024)

MechRL: Reinforcement Learning Agents Perform Circuit Discovery for Mechanistic Interpretability
by: Khadka, Barsat
Published: (2026)

Recurrent-Depth VLA: Implicit Test-Time Compute Scaling of Vision-Language-Action Models via Latent Iterative Reasoning
by: Tur, Yalcin, et al.
Published: (2026)

Generate Any Scene: Scene Graph Driven Data Synthesis for Visual Generation Training
by: Gao, Ziqi, et al.
Published: (2024)