Machine learning is a vital part of many real-world systems, but several concerns remain about the lack of interpretability, explainability and robustness of black-box AI systems. Concept Bottleneck Models (CBM) address some of these challenges by learning interpretable concepts from high-dimensional data, e.g. images, which are used to predict labels. In this talk, we describe a framework that provides a statistical estimator that provable extracts concepts from data. Our framework leverages causal representation learning (CRL) methods to learn latent causal variables from high-dimensional observations in an unsupervised way, and then learns to align these variables with interpretable concepts with only few concept labels. We propose a linear and a non-parametric estimator for this mapping. We evaluate our framework in synthetic and image benchmarks, showing that the learned concepts have less impurities and are often more accurate than other CBMs, even in settings with strong correlations between concepts. Finally, we will describe how this method could potentially be deployed in settings such as fraud detection and anomaly detection.

Hidde is a finishing PhD student in mathematical machine learning at the Korteweg-de Vries institute at the University of Amsterdam. His research is focused on developing mathematical foundations of explainabile AI and interpretable machine learning methods, under supervision of dr. Tim van Erven. He has gained hands on experience by developing data-driven and machine learning powered solutions for companies such as Booking.com and Amsterdam Data Collective.