Teaching

In the realm of biomedical research, graph deep learning (DL) has emerged as a pivotal analytical tool. Graph neural networks (GNNs) have demonstrated remarkable power in analyzing intricate biological datasets such as molecular graphs, protein-protein interaction networks, and patient similarity networks. Despite their efficacy, these sophisticated models often resemble enigmatic black boxes, with millions of parameters obscuring their decision-making processes. In life-critical fields like biomedicine, where understanding the 'why' behind model predictions is just as crucial as the predictions themselves, the challenge of explainability becomes paramount. This tutorial equips participants with the knowledge and tools to tackle this challenge — covering graph DL in biomedicine, core principles of explainability in GNNs, practical biomedical applications, interpretable GNN models, and hands-on exercises.

As predictive models are increasingly deployed in critical domains — healthcare, law, and finance — there is growing emphasis on explaining model predictions to decision makers so they can understand the rationale and determine when to rely on these predictions. This tutorial covers eXplainable Artificial Intelligence (XAI) algorithms that generate explanations of individual predictions made by complex ML models, focusing on feature attribution methods, their applications in high-stakes domains, and ensuring that explanations are reliable and credible for relevant stakeholders.

Machine learning is increasingly deployed in high-stakes decision-making. Explainable ML (XAI) is often pitched as a panacea for managing uncertainty and skepticism. While technical limitations of explainability methods are being characterized formally in ML literature, the impact of explanation methods and their limitations on end users — policy makers, judges, doctors — is not well understood. This tutorial contextualizes explanation methods and their limitations for such end users, discussing overarching ethical implications of technical challenges beyond misleading decision-making, with applications in finance, clinical healthcare, and criminal justice.