LINE: LLM-based Iterative Neuron Explanations for Vision Models
2026-04-09 • Computer Vision and Pattern Recognition
Computer Vision and Pattern RecognitionArtificial IntelligenceMachine Learning
AI summaryⓘ
The authors present LINE, a new method to understand what individual neurons in vision AI systems represent by naming concepts without needing training or fixed word lists. LINE uses a language model and image generator together, guessing and improving concept ideas based on neuron activity, fully treating the AI system as a black box. Their method works better than previous ones on popular image datasets and finds many new concepts that old methods miss. It also helps explain neurons that respond to multiple concepts and provides visual examples to support its findings.
deep neural networksneuron labelingconcept interpretabilityblack-box settinglarge language modeltext-to-image generationImageNetactivation maximizationpolysemanticityvision models
Authors
Vladimir Zaigrajew, Michał Piechota, Gaspar Sekula, Przemysław Biecek
Abstract
Interpreting the concepts encoded by individual neurons in deep neural networks is a crucial step towards understanding their complex decision-making processes and ensuring AI safety. Despite recent progress in neuron labeling, existing methods often limit the search space to predefined concept vocabularies or produce overly specific descriptions that fail to capture higher-order, global concepts. We introduce LINE, a novel, training-free iterative approach tailored for open-vocabulary concept labeling in vision models. Operating in a strictly black-box setting, LINE leverages a large language model and a text-to-image generator to iteratively propose and refine concepts in a closed loop, guided by activation history. We demonstrate that LINE achieves state-of-the-art performance across multiple model architectures, yielding AUC improvements of up to 0.18 on ImageNet and 0.05 on Places365, while discovering, on average, 29% of new concepts missed by massive predefined vocabularies. Beyond identifying the top concept, LINE provides a complete generation history, which enables polysemanticity evaluation and produces supporting visual explanations that rival gradient-dependent activation maximization methods.