CB-SLICE: Concept-Based Interpretable Error Slice Discovery

2026-05-28

AI Summary

Deep learning models often exhibit systematic errors—referred to as error slices—on specific subgroups, yet existing approaches struggle to precisely identify their root causes. This work proposes a novel method based on Concept Bottleneck Models (CBMs) that, for the first time, leverages failures in predicting human-interpretable semantic concepts to directly discover and explain error slices. By integrating concept representations with keyword identification, the approach clusters samples sharing common concept-level errors and performs attribution analysis, yielding fine-grained, high-fidelity explanations tightly aligned with the model’s reasoning process. Experiments across multiple benchmarks demonstrate that the method outperforms state-of-the-art alternatives, not only uncovering known biases more accurately but also generating richer and more faithful error attributions.

Abstract

Despite strong average-case performance, deep learning models often exhibit systematic errors on specific population groups, known as error slices. Identifying these groups and the root causes of their failures is critical for model debugging and bias mitigation. However, existing error Slice Discovery Methods (SDMs) typically generate explanations disconnected from the model's inference process, thus only approximating the underlying error source and may be inaccurate. We address this limitation by leveraging Concept Bottleneck Models (CBMs), whose predictions are directly dependent on human-understandable semantic concepts. Since downstream task failures in CBMs commonly arise from concept mispredictions, concept representations provide a strong candidate for error slice identification, offering fine-grained explanations directly linked to the error source. Building on this insight, we introduce CB-SLICE, a concept-based SDM that groups samples with shared concept prediction failures and identifies the keyword concepts most responsible for each slice's failure mode. Across multiple benchmarks, we show that CB-SLICE outperforms state-of-the-art methods in uncovering well-known biases while providing richer and more faithful explanations of model errors.

Problem

• error slice discovery
• model interpretability
• systematic errors
• bias mitigation
• concept-based explanation

Innovation

• Concept Bottleneck Models
• Error Slice Discovery
• Interpretable AI
• Model Debugging
• Bias Mitigation