DictXAI improves AI explanations by linking predictions to input concepts
Revisiting Explainable AI through Model-Independent Concept Dictionaries
Machine Learning
Summary
Complex AI models are hard to understand, especially when their explanations rely on unclear or model-specific features. The authors introduce DictXAI, a method that breaks down inputs into understandable pieces using a dictionary of meaningful elements. This lets the method show exactly which parts of the input led to a prediction, helping spot errors or biases. Their approach works across different models and types of data, making AI explanations clearer and easier to trust.
What this means in practice
- •For medical ai engineers: Identify which specific biomedical signal patterns contribute to AI model predictions to improve diagnostic tools.
- •For image analysis developers: Detect artifact patterns causing AI model errors by linking outputs directly to interpretable image elements.
Authors
Thomas Schnake, Doreen Schöppenthau, Alexander Meyer, Jacques Corbeil, Klaus-Robert Müller, Grégoire Montavon
Abstract
Modern applications of AI rely on increasingly complex models. Explainable AI (XAI) has emerged as a set of techniques aimed at improving model transparency. However, existing XAI methods typically assume input features to be inherently interpretable, or they rely on intermediate internal abstractions that are difficult to characterize and highly architecture-specific, hindering consistent use across models. To address these limitations, we propose DictXAI, a method that defines concepts directly in the input domain via a dictionary---a large, potentially overcomplete set of predefined elements, each carrying an interpretable meaning. Technically, DictXAI first computes a sparse code of the input and then attributes the model's prediction to the associated dictionary elements. We demonstrate the actionable nature of DictXAI explanations, showing that they can attribute AI malfunctions (e.g., Clever Hans effects) directly to identifiable artifact patterns in the data, while fostering human-AI alignment on intricate biomedical signals. We further demonstrate our method's ability to operate across a wide variety of dictionaries, including learned image bases, analytically defined waveforms for electrocardiography, and experimentally acquired dictionary elements. Overall, our results show that DictXAI provides more interpretable, actionable, and architecture-agnostic insights than classical XAI or existing concept-based approaches.