GlanceNets: Interpretabile, Leak-proof Concept-based Models
This addresses the need for more reliable interpretability in AI models, particularly for domains requiring transparent decision-making, though it is incremental in improving existing concept-based methods.
The paper tackles the problem of interpretability in concept-based models by defining interpretability as alignment with the underlying data generation process and introduces GlanceNets, which achieves better alignment than state-of-the-art approaches while preventing spurious information leakage.
There is growing interest in concept-based models (CBMs) that combine high-performance and interpretability by acquiring and reasoning with a vocabulary of high-level concepts. A key requirement is that the concepts be interpretable. Existing CBMs tackle this desideratum using a variety of heuristics based on unclear notions of interpretability, and fail to acquire concepts with the intended semantics. We address this by providing a clear definition of interpretability in terms of alignment between the model's representation and an underlying data generation process, and introduce GlanceNets, a new CBM that exploits techniques from disentangled representation learning and open-set recognition to achieve alignment, thus improving the interpretability of the learned concepts. We show that GlanceNets, paired with concept-level supervision, achieve better alignment than state-of-the-art approaches while preventing spurious information from unintendedly leaking into the learned concepts.