The lack of transparency in AI classification models remains a persistent challenge in domains where the reasoning behind a prediction matters, such as finance, healthcare, and legal applications. Researchers at the University of Amsterdam present ECSEL at ICML 2026, an approach that learns interpretable models explaining their decisions in the form of mathematical equations.
ECSEL (Explainable Classification via Signomial Equation Learning) was developed by Adia Lumadjeng, Ilker Birbil, and Erman Acar in response to the growing demand for explainable artificial intelligence. The method learns signomial equations that provide insight into how a classification model reaches an outcome, in a form that is understandable to people outside the model.
The approach is distinctive in that explanations are not generated after the fact, but are part of the learning process itself. This makes ECSEL an alternative to black-box models in situations where transparency is legally required or operationally necessary, for example in credit decisions, fraud detection, or medical diagnostics. The paper was presented at the International Conference on Machine Learning (ICML 2026).
Read also
