Semantic Prototypes: Enhancing Transparency Without Black Boxes
Orfeas Menis-Mastromichalakis, Giorgos Filandrianos, Jason Liartis, Edmund Dervakos, Giorgos Stamou
TL;DR
The paper tackles the explainability gap in complex ML systems by replacing opaque, latent-space or raw-feature prototypes with semantically grounded prototypes built from Attribute Set Descriptions (ASD) and Class Cluster Descriptions (CCD). It develops a two-stage method: (i) mining CCDs to semantically describe class clusters via a greedy, set-cover-like process using a similarity metric over ASDs, and (ii) selecting semantic prototypes as the closest data points to CCDs based on a set-edit distance, accompanied by semantically meaningful explanations. Empirical results in CLEVR-Hans and CUB-200, plus a human user survey, show that semantic prototypes improve interpretability and human understanding, with ProtoSem achieving the highest accuracy and user preference among baselines. The work highlights the practical impact of presenting explanations at a semantic level, enabling clearer insights and trust, and outlines directions for extending the framework with LLMs, knowledge graphs, and broader domain applications.
Abstract
As machine learning (ML) models and datasets increase in complexity, the demand for methods that enhance explainability and interpretability becomes paramount. Prototypes, by encapsulating essential characteristics within data, offer insights that enable tactical decision-making and enhance transparency. Traditional prototype methods often rely on sub-symbolic raw data and opaque latent spaces, reducing explainability and increasing the risk of misinterpretations. This paper presents a novel framework that utilizes semantic descriptions to define prototypes and provide clear explanations, effectively addressing the shortcomings of conventional methods. Our approach leverages concept-based descriptions to cluster data on the semantic level, ensuring that prototypes not only represent underlying properties intuitively but are also straightforward to interpret. Our method simplifies the interpretative process and effectively bridges the gap between complex data structures and human cognitive processes, thereby enhancing transparency and fostering trust. Our approach outperforms existing widely-used prototype methods in facilitating human understanding and informativeness, as validated through a user survey.