Data Science Seminars - Overcoming Medical Data Barriers with Generative AI and Federated Learning
Wednesday, June 25, 2025 | 5:00 p.m.
Data Science and Bioinformatics Lab (Building 21)
Dipartimento di Elettronica, Informazione e Bioingegneria - Politecnico di Milano
Speaker: Francesca Pia Panaccione (Politecnico di Milano)
Contacts: Silvia Cascianelli | silvia.cascianelli@polimi.it
Data Science and Bioinformatics Lab (Building 21)
Dipartimento di Elettronica, Informazione e Bioingegneria - Politecnico di Milano
Speaker: Francesca Pia Panaccione (Politecnico di Milano)
Contacts: Silvia Cascianelli | silvia.cascianelli@polimi.it
Abstract
Wednesday, June 25, 2025 at 5:00 p.m. Francesca Pia Panaccione (Politecnico di Milano) will hold a seminar titled "Concept Activation Vectors: Probing Human-understandable Concepts in Image Classification Networks" in the Data Science and Bioinformatics Lab (Building 21).
The event is part of the Data Science Seminars organized by the Data Science Lab at Politecnico di Milano.
Advancements in healthcare AI depend on access to large-scale, high-quality datasets. However, in practice, medical data is often scarce, fragmented, and difficult to share. This scarcity is particularly pronounced in rare diseases and novel clinical domains, where data collection and annotation are costly and time-consuming. At the same time, strict privacy regulations—such as GDPR—pose significant barriers to data sharing and cross-institutional collaboration.
These challenges limit model generalizability and slow progress in precision medicine. To address them, complementary strategies have emerged. Among them, Generative AI offers a way to create realistic, privacy-preserving synthetic data, enabling robust model training even in data-poor settings. Federated learning, on the other hand, supports decentralized model development across institutions, allowing for collaborative training without disclosing sensitive patient data.
In this seminar, we will explore how generative AI and federated learning can help mitigate privacy-related issues in medical data analysis. In particular, we will focus on two use cases: the design of a multimodal generative model for synthesizing gene expression data from clinical data and histopathological images, and an application of federated learning to enable secure AI collaboration among medical institutions across Europe.
The event is part of the Data Science Seminars organized by the Data Science Lab at Politecnico di Milano.
Advancements in healthcare AI depend on access to large-scale, high-quality datasets. However, in practice, medical data is often scarce, fragmented, and difficult to share. This scarcity is particularly pronounced in rare diseases and novel clinical domains, where data collection and annotation are costly and time-consuming. At the same time, strict privacy regulations—such as GDPR—pose significant barriers to data sharing and cross-institutional collaboration.
These challenges limit model generalizability and slow progress in precision medicine. To address them, complementary strategies have emerged. Among them, Generative AI offers a way to create realistic, privacy-preserving synthetic data, enabling robust model training even in data-poor settings. Federated learning, on the other hand, supports decentralized model development across institutions, allowing for collaborative training without disclosing sensitive patient data.
In this seminar, we will explore how generative AI and federated learning can help mitigate privacy-related issues in medical data analysis. In particular, we will focus on two use cases: the design of a multimodal generative model for synthesizing gene expression data from clinical data and histopathological images, and an application of federated learning to enable secure AI collaboration among medical institutions across Europe.
