The paper investigates how Concept Activation Vectors (CAVs), used for concept‑based explanations in AI, vary due to the randomness in sampling training examples. A theoretical analysis shows that the variance of CAVs decreases roughly as 1/N, where N is the number of sampled examples. Experiments across four modalities confirm this relationship, leading to practical guidelines for efficiently applying CAVs.
arXiv:2605.15688v2 Announce Type: replace-cross Abstract: Concept Activation Vectors (CAV) and the Testing with CAVs (TCAV) method are fundamental tools for concept-based explainability in deep learn...
arXiv:2606. 16535v1 Announce Type: new Abstract: Concept Bottleneck Models (CBMs) are a relevant tool for explainable Artificial Intelligence because they make their predictions through human-interpretable symbols.
The paper introduces Ra‑NEM, a method that optimizes the area under insertion and deletion curves to improve the faithfulness of feature importance attributions in machine learning models. By deriving an objective function from faithfulness and approximating its gradient, the authors link insertion curves to top‑k feature selection, enabling a loss function that measures attribution quality. Ra‑NEM, integrated with the neural explanation mask framework, works with any differentiable model, yielding accurate, robust, and fast attributions that outperform existing algorithms on multiple XAI metrics.
arXiv:2607. 04222v1 Announce Type: new Abstract: Interpretability methods aim to reveal the features represented inside large language models (LLMs).
arXiv:2411. 05698v3 Announce Type: replace-cross Abstract: Convolutional Neural Networks (CNNs) have shown remarkable performance in image classification.
The paper introduces ICON decomposition, a method for auditing neural networks by decomposing the variance of intermediate representations into contributions from Independent Canonical Concepts (ICON). Unlike traditional decodability metrics, ICON isolates each concept’s influence while conditioning on all other concepts and the outcome, providing a clearer view of which concepts truly drive predictions. Experiments on simulated data and real-world skin‑cancer and neuroimaging tasks show ICON outperforms seven baseline methods and successfully distinguishes genuine shortcuts from merely correlated features.
The paper introduces a method for efficiently exploring the Rashomon set of Concept Bottleneck Models (CBMs) by using a parallel parameter‑efficient adaptation module, checkpointing, and a concept diversity objective. This approach generates multiple equally accurate CBMs from a single training process, achieving greater diversity than baseline methods while consuming less memory. The resulting diverse models enable trustworthy selection, reduce inter‑class confusion, and support reliable abstention in decision‑making.
The paper introduces a concept-guided poisoning unlearning framework that explicitly identifies semantic concepts most strongly linked to a target class or knowledge. For image classification, it uses a Post-hoc Concept Bottleneck Model to locate relevant image regions and creates poisoned samples; for large language models, it aggregates Integrated Gradients over multiple question responses to mask important content and generate poisoned training targets. Experiments across several datasets demonstrate effective unlearning while largely preserving model utility.
arXiv:2605.25903v2 Announce Type: replace-cross Abstract: Activation verbalization explains hidden representations in natural language, but existing methods are mostly limited to self-explanation, wh...
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.
arXiv:2603. 14894v3 Announce Type: replace-cross Abstract: Trust and ethical concerns due to the widespread deployment of opaque machine learning (ML) models motivating the need for reliable model explanations.
arXiv:2607. 04548v1 Announce Type: cross Abstract: Novel category discovery aims to identify unseen classes from unlabeled data by transferring knowledge from labeled categories, but most existing methods perform discovery in opaque latent feature spaces.