MLLMs show promise for low-cost concept-based AI explanations

Researchers have developed a training-free approach for generating localized explanations in Explainable AI (XAI) using Multimodal Large Language Models (MLLMs). Their method, called Concept Naming (CoNa), evaluates how well these models can identify semantic concepts within specific regions of images, even at the object and part levels. Experiments with MLLMs ranging from 7B to 32B parameters demonstrated significant accuracy in object-level concept naming, suggesting a path towards more cost-effective XAI research. AI

IMPACT This research could lead to more accessible and cost-effective methods for understanding AI model decisions.

RANK_REASON The item is a research paper published on arXiv detailing a new method for XAI. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.AI →

• 7B
• arXiv
• CoNa
• Multimodal Large Language Models
• Open-CoNa

• paper
• other

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →