Multimodal Foundation Models and Agents for Biological Discovery and Personalized Therapeutics

Cancelled
Event details
Date | 22.07.2025 |
Hour | 10:30 › 12:30 |
Speaker | Eeshaan Jain |
Location | |
Category | Conferences - Seminars |
EDIC candidacy exam
Exam president: Prof. Martin Jaggi
Thesis advisor: Prof. Charlotte Bunne
Co-examiner: Dr Dorina Thanou
Abstract
This research aims to develop machine learning methods that enable robust, efficient, and personalized decision-making in biomedical settings. The work focuses on three interconnected directions. First, I will develop foundation models for high-dimensional tissue dataâsuch as histopathology and multiplexed imagingâthat learn unified representations by integrating spatial, morphological, and molecular signals. These models are designed to generalize across data modalities and serve as a basis for downstream tasks. Second, I will explore test-time adaptation under constrained settings, where only partial input is available. By learning strategies for sequential feature acquisition, the goal is to maintain diagnostic accuracy while minimizing cost and information redundancy. Third, I will extend these methods toward the development of AI agents capable of structured reasoning across heterogeneous data sources. These agents will support personalized therapeutic recommendations by combining model outputs with contextual and external knowledge. Together, this work contributes toward building adaptive, scalable, and clinically relevant AI systems for precision medicine.
Selected papers
coming soon
Exam president: Prof. Martin Jaggi
Thesis advisor: Prof. Charlotte Bunne
Co-examiner: Dr Dorina Thanou
Abstract
This research aims to develop machine learning methods that enable robust, efficient, and personalized decision-making in biomedical settings. The work focuses on three interconnected directions. First, I will develop foundation models for high-dimensional tissue dataâsuch as histopathology and multiplexed imagingâthat learn unified representations by integrating spatial, morphological, and molecular signals. These models are designed to generalize across data modalities and serve as a basis for downstream tasks. Second, I will explore test-time adaptation under constrained settings, where only partial input is available. By learning strategies for sequential feature acquisition, the goal is to maintain diagnostic accuracy while minimizing cost and information redundancy. Third, I will extend these methods toward the development of AI agents capable of structured reasoning across heterogeneous data sources. These agents will support personalized therapeutic recommendations by combining model outputs with contextual and external knowledge. Together, this work contributes toward building adaptive, scalable, and clinically relevant AI systems for precision medicine.
Selected papers
coming soon
Practical information
- General public
- Free
Contact
- edic@epfl.ch