Summary
This paper introduces concept-based abductive and contrastive explanations for deep neural networks, combining two research threads: concept-based explanations that describe predictions in high-level, human-understandable terms, and formal abductive/contrastive explanations that identify minimal sets of causally relevant input features (previously limited to low-level features like pixels). The authors—Ronaldo Canizales, Divya Gopinath, Corina Păsăreanu, and Ravi Mangal—propose explanations that capture minimal sets of high-level concepts causally connected to model outputs. They present a family of algorithms to enumerate all minimal explanations, using concept-erasure procedures to establish causal links. By aggregating these explanations, the method explains not only predictions on individual images but also model behavior across sets of images sharing user-specified common behaviors. Evaluation across multiple models, datasets, and behaviors demonstrates the effectiveness of computing helpful, user-friendly explanations. Preprint available on arXiv: 2605.06640.
Paper Overview
Field: Machine Learning
Authors: Ronaldo Canizales, Divya Gopinath, Corina Păsăreanu, Ravi Mangal
Published: 2026-05-07
arXiv: 2605.06640
Abstract
Concept-based explanations offer a promising way to explain deep neural network predictions in terms of high-level, human-understandable concepts. However, existing approaches either fail to establish a causal connection between concepts and model predictions, or have limited expressiveness, inferring causal explanations involving only a single concept.
In parallel, work on formal abductive and contrastive explanations computes minimal sets of input features causally relevant to model outcomes, but considers only low-level features such as pixels.
Merging these two lines of research, this work introduces the notion of concept-based abductive and contrastive explanations, which capture minimal sets of high-level concepts causally relevant to model outcomes. The authors then propose a family of algorithms to enumerate all minimal explanations, leveraging concept-erasure procedures to establish causality.
By appropriately aggregating such explanations, the approach can explain:
- Model predictions on individual images
- Model behavior across sets of images exhibiting user-specified common behaviors
The method was evaluated on multiple models, datasets, and behaviors, demonstrating its effectiveness in computing helpful, user-friendly explanations.
---
*Auto-collected on 2026-05-10*
This page is an English static mirror generated for search and AI citation.
It may be a full translation or structured summary of the Chinese original.
Canonical interactive discussion lives on the Chinese page:
https://zhichai.net/topic/177619701