Research map / Label-efficient and robust learning / Recognition and long-tail
IDEA: Unwrapping Visual Black-box Models by Interaction Decomposition
Chaojie Ji, Jian Xu, He Zhang, Hongyan Wu, Yankai Cao, Ruxin Wang
Read the paperCheck your idea against 137,217 papers
AI-extracted summary. Written by AI from the paper's abstract only. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Proposes IDEA, a concept-based explanation method that decomposes a visual black-box model's predictions to capture interactions between a concept and its visual context, rather than assigning a scalar importance score.
Problem
Existing concept-based explanation methods for deep vision models assign only a scalar importance score to each concept and ignore how the concept interacts with its visual context, leaving predictions only partially explained.
Approach
Decomposes the model's prediction into concept-context interaction terms so that explanations reflect how a concept and its surrounding visual context jointly drive the output, instead of a single scalar attribution per concept.
Closest papers by meaning
- Overlooked Factors in Concept-Based Explanations: Dataset Choice, Concept Learnability, and Human Capability (CVPR 2023 · similarity 0.90)
- Attention-based Interpretability with Concept Transformers (ICLR 2022 · similarity 0.89)
- FaCT: Faithful Concept Traces for Explaining Neural Network Decisions (NeurIPS 2025 · similarity 0.89)
- CoE: Chain-of-Explanation via Automatic Visual Concept Circuit Description and Polysemanticity Quantification (CVPR 2025 · similarity 0.88)
- Understanding Inter-Concept Relationships in Concept-Based Models (ICML 2024 · similarity 0.88)
- Concept-RuleNet: Grounded Multi-Agent Neurosymbolic Reasoning in Vision Language Models (AAAI 2026 · similarity 0.88)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
