Where and when

AIAI 2024 is a workshop organized by members of the National Group for Scientific Computing (GNCS), associated with both the Department of Computer Science and the Department of Mathematics at the University of Bari "Aldo Moro". The workshop will focus on examining how advancements and research in new numerical strategies influence artificial intelligence algorithms, enhancing both their accuracy and efficiency, and improving traditional approaches. Some key topics of interest include numerical optimization strategies, numerical linear algebra, deterministic and randomized algorithms, machine learning, and artificial intelligence. The extensive application of artificial intelligence across various fields, from academic research to industry, will be a major point of discussion in a roundtable organized in collaboration with CIRPAS (Interuniversity Center for Research on "Population, Environment, Health"), which will explore the social impact and implications of AI. _{^{[source: https://ai2workshop.uniba.it/wordpress/]}}

Program of the workshop

The MIDA group will take part at the workshop with a talk

Title: Integrating data-driven and human-driven insights in generative disease progression modeling. 

Speaker: Sara Garbarino

Abstract: Disease progression models are a family of statistical and machine learning tools that have been developed to learn long-term disease biomarker timelines for chronic diseases from short-term data without requiring prior knowledge of an individual’s disease stage. Such models enable biomarker changes to be mapped at a fine-grained temporal resolution. We place disease progression modelling within a unified general framework that combines generative learning with constraints informed by human or biological insights to infer both the data-driven time axis and the shape of biomarker trajectories along it. The data-driven time axis represents a model-based time axis that describes the expected average disease progression, enabling the temporal realignment of individuals relative to this timeline. This approach allows short-term data to inform long-term disease tra- jectories, whereas a set of trajectory constraints informed by human insights enable reconstruction of trajectories from noisy medical datasets and ensure interpretability of outputs. In contrast to ‘black box’ machine learning tools, data-driven disease progression models typically require fewer data and are inherently interpretable, thereby aiding disease understanding in addition to enabling classification, prediction and stratification. We specifically focus on graph-based disease progression models and propose an integrated approach that combines dynamical system modeling on graphs with generative learning to model the propagation of pathological proteins in the brains of individuals with Alzheimer’s disease.