LINE: LLM-based Iterative Neuron Explanations for Vision Models

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Zaigrajew, Vladimir, Piechota, Michał, Sekula, Gaspar, Gelar, Paweł, Biecek, Przemysław
Format: Preprint
Published: 2026
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866916006231801856
author Zaigrajew, Vladimir
Piechota, Michał
Sekula, Gaspar
Gelar, Paweł
Biecek, Przemysław
author_facet Zaigrajew, Vladimir
Piechota, Michał
Sekula, Gaspar
Gelar, Paweł
Biecek, Przemysław
contents Interpreting 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. LINE achieves state-of-the-art performance across multiple model architectures, yielding AUC improvements of up to 0.11 on ImageNet and 0.05 on Places365, while discovering, on average, 27% of new concepts missed by predefined vocabularies. Beyond identifying the top concept, LINE provides a complete generation history, enabling polysemanticity evaluation and producing visual explanations that rival gradient-dependent activation maximization methods. The source code will be made available soon.
format Preprint
id arxiv_https___arxiv_org_abs_2604_08039
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle LINE: LLM-based Iterative Neuron Explanations for Vision Models
Zaigrajew, Vladimir
Piechota, Michał
Sekula, Gaspar
Gelar, Paweł
Biecek, Przemysław
Computer Vision and Pattern Recognition
Artificial Intelligence
Machine Learning
Interpreting 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. LINE achieves state-of-the-art performance across multiple model architectures, yielding AUC improvements of up to 0.11 on ImageNet and 0.05 on Places365, while discovering, on average, 27% of new concepts missed by predefined vocabularies. Beyond identifying the top concept, LINE provides a complete generation history, enabling polysemanticity evaluation and producing visual explanations that rival gradient-dependent activation maximization methods. The source code will be made available soon.
title LINE: LLM-based Iterative Neuron Explanations for Vision Models
topic Computer Vision and Pattern Recognition
Artificial Intelligence
Machine Learning
url https://arxiv.org/abs/2604.08039