The paper introduces ICON Decomposition, a method for auditing deep neural networks by decomposing layer-wise representations into independent concept contributions. Unlike existing techniques that rely on linear probes or concept activation vectors, ICON quantifies the variance share each concept explains while conditioning on all other concepts and the outcome, allowing comparison across layers and concept types. Experiments on simulated data, skin‑cancer, and neuroimaging models show that ICON more accurately recovers true concept importance and can distinguish learned shortcuts from correlated concepts, as validated by retraining and out‑of‑distribution tests.
By Roshan Prakash Rane, Marco Simnacher, Manuel Pfeuffer, Marc-Andre Schulz, Nys Tjade Siegel, Maximilian Dreyer, Frederik Pahde, Wojciech Samek, Sonja Greven, Kerstin Ritter
arXiv:2608.26083v2 Announce Type: replace-cross
Abstract: Deep neural networks often exploit spurious associations, a failure known as shortcut learning. Auditing for shortcuts requires testing many...
By Roshan Prakash Rane, Marco Simnacher, Manuel Pfeuffer, Marc-Andre Schulz, Nys Tjade Siegel, Maximilian Dreyer, Frederik Pahde, Wojciech Samek, Sonja Greven, Kerstin Ritter
ICON Decomposition is a new method for explaining deep neural networks by quantifying how much variance each concept explains in a network layer after accounting for all other concepts and the outcome. Unlike previous concept‑based methods that evaluate concepts in isolation, ICON can distinguish genuine model reliance from spurious correlations. Experiments on synthetic data, skin‑lesion, and brain‑imaging models show that ICON recovers concept importance more accurately, isolates truly relied‑upon concepts, and provides sparse explanations validated through retraining and out‑of‑distribution testing.
By Roshan Prakash Rane, Marco Simnacher, Manuel Pfeuffer, Marc-Andre Schulz, Nys Tjade Siegel, Maximilian Dreyer, Frederik Pahde, Wojciech Samek, Sonja Greven, Kerstin Ritter
arXiv:2510. 15770v4 Announce Type: replace-cross Abstract: Although deep neural networks achieve strong predictive performance, their internal reasoning often remains difficult to inspect and control.
By Gaoxiang Huang, Songning Lai, Yutao Yue
arXiv:2607. 09649v1 Announce Type: new Abstract: Concept-based explainable artificial intelligence (AI) can make model reasoning more human-understandable, but concept-level outputs are not automatically trustworthy.
By Mohadeseh Mollapour, Koorosh Aslansefat, Zeinab Dehghani, Bhupesh Kumar Mishra, Tejal Shah, Zhibao Mian
arXiv:2606. 01843v1 Announce Type: cross Abstract: Deepfake detection suffers from poor generalization across forgery methods, as existing models tend to rely on spurious method-specific shortcuts that fail to transfer to unseen manipulations.
By Yihui Wang, Yonghui Yang, Jilong Liu, Fengbin Zhu, Le Wu, Tat-Seng Chua
arXiv:2606. 30498v1 Announce Type: cross Abstract: Human decision-making interprets the world through high-level concepts, such as recognizing a bird by its belly color.
By Laines Schmalwasser, Jan Blunk, Niklas Penzel, Julia Niebling, Joachim Denzler
The paper introduces AVCE, a method for verifiable concept erasure in diffusion models that audits latent representations to identify vulnerable directions and then edits cross‑attention and self‑attention projections in closed form. By fine‑tuning with pathway‑level auditing losses and orthogonal gradient projection, AVCE reduces attack success rates by 5.07× and improves auditing scores by 3.84× compared to the best baseline while maintaining generation quality.
By Kaiyuan Deng, Yuchen Li, Gen Li, Yang Xiao, Geng Yuan, Xiaoyong Yuan, Bo Hui, Xiaolong Ma
arXiv:2601. 21944v3 Announce Type: replace Abstract: The widespread adoption of deep learning models in computer vision has intensified concerns about interpretability.
By Konstantinos P. Panousis, Diego Marcos
The paper presents a plug‑and‑play framework that uses concept-based interpretation to debug medical imaging models. By aligning a single‑modality encoder with BioMedCLIP, the authors build a Concept Bottleneck Model that allows concept‑level interventions to distinguish causal from spurious features. Experiments on Mayo Clinic ultrasound and CheXpert chest X‑ray data show that these interventions can reliably diagnose model issues and even improve performance through guided fine‑tuning.
By Samrajya Thapa, Daniel J. Quest, Timothy L. Kline, Carrie L. Langstraat, Emanuel C. Trabuco, Wei Le
The paper investigates whether neural networks exhibit conceptual separation, meaning that examples of the same concept cluster together and related concepts are closer in representation space. Using geometric and distributional analyses, the authors find that Convolutional Neural Networks (CNNs) produce coherent, semantically ordered representations for familiar ImageNet concepts, but this coherence weakens for unseen concepts and under domain shift. Large Language Models (LLMs) keep distinct domains well separated, bring related subdomains closer, yet lose distinction between ambiguous topics at both mean and covariance levels.
By Jaee Ponde, Roshni Agarwal, Subhashis Banerjee
arXiv:2608. 15731v1 Announce Type: cross Abstract: Deep Neural Networks (DNNs) deployed in high-risk domains, such as healthcare and autonomous driving, must be not only accurate but also understandable to ensure user trust.
By Haadia Amjad, Ronald Tetzlaff