What Breaks in Multi-Image Schema Application? Factorial Oracle Interventions Reveal an Instance-Information Bottleneck
cs.CV
Submitted: 2026-10-14
Updated: 2026-10-14
License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
Sources
- Benchmarking Multi-Image Understanding in Vision and Language Models: Perception, Knowledge, Reasoning, and Multi-Hop Reasoning
- Needle In A Multimodal Haystack
- Image First or Text First? Optimising the Sequencing of Modalities in Large Language Model Prompting and Reasoning Tasks
- DISSECT: Diagnosing Where Vision Ends and Language Priors Begin in Scientific VLMs
- Perceptual Taxonomy: Evaluating and Guiding Hierarchical Scene Reasoning in Vision-Language Models
- Disentangling Perception and Reasoning in Multimodal LLMs via Reward Design
- SCRIBE: Structured Mid-Level Supervision for Tool-Using Language Models
- OSEF: One-Step Evidence Fusion for Cross-Video Scene Procedure Planning
- Adaptive Information Control for Search-Augmented LLM Reasoning
- How Early Is Early Enough? Design-Dependent Observation-Window Sufficiency in Subscription Churn Prediction
- Validation-Stage Combinatorial Fusion Analysis for Imbalanced Credit-Card Fraud Detection
- Interpretable vs Learned Encoders for High-Cardinality Fraud Detection
- Interpretable Factor Decomposition for Decision Intelligence in Large-Scale Financial Markets: Evidence from China's A-Share Market
- Decomposing Firm-Level Crisis Responses from Incomplete Market Signals: Evidence from China's IT Sector During COVID-19
- Explainable Machine Learning for Broadband Adoption Disparities: Tract-Level Prediction and SHAP-Based Factor Profiling
- Support Thresholds, Not Algorithms, Limit Rare-Association Recovery in Co-Purchase Networks
- MAVEN-T: Reinforced Heterogeneous Distillation for Real-Time Multi-Agent Trajectory Prediction
- RACER: Reinforced Agent Collaboration for Explainable Reasoning on Knowledge Graphs
- From Agent Traces to Trust: A Survey of Evidence Tracing and Execution Provenance in LLM Agents
Related papers
- Loss Knows Best: Detecting Annotation Errors in Videos via Loss Trajectories
- AnchorWeave: World-Consistent Video Generation with Retrieved Local Spatial Memories
- Benchmarking the Robustness of Foundation Models for Mammography under Domain Shift
- MambaX-Net: Dual-Input Mamba-Enhanced Cross-Attention Network for Longitudinal MRI Segmentation
- TeleOCR: Navigating Document Parsing Across Digital and Camera-Captured Documents
- A Survey on Efficient Vision-Language-Action Models