Interpretability has been a central theme of CILab research since well before the recent emergence of Explainable Artificial Intelligence. Our work in this area originates from fuzzy systems and granular computing, where the representation of uncertain, imprecise and human-understandable concepts is a fundamental part of model design.
We investigate fuzzy and granular models for representing and reasoning with complex information, including rule-based modeling, information granulation, fuzzy clustering and neuro-fuzzy approaches. Particular attention has traditionally been devoted to balancing predictive capability with interpretability and to designing models whose internal representations can be meaningfully inspected by humans.
Another established research direction concerns adaptive learning from evolving data. Fuzzy and incremental learning methods have been investigated for data streams and dynamic environments, where models need to adapt as new information becomes available and data distributions change over time.
This research naturally connects with modern Explainable AI (XAI). We investigate both intrinsically interpretable models and methods for explaining the predictions of machine learning and deep learning systems, together with methodologies for assessing the quality, faithfulness and usefulness of their explanations.
In this way, contemporary research on Explainable AI extends a long-standing CILab tradition in interpretable Computational Intelligence while connecting it with modern machine learning and deep learning.