About the Speaker
Guillaume Jaume is a tenure-track Assistant Professor in the Department of Fundamental Oncology at the University of Lausanne, where he leads a lab on AI for computational pathology and oncology. His research spans foundation models for histopathology, vision-language models, and the integration of tissue imaging with spatial omics. He holds an ERC Starting Grant and is co-founder of Tremont AI, which develops AI solutions for preclinical drug safety. He was previously a postdoctoral researcher at Harvard Medical School and Brigham and Women’s Hospital, and earned his PhD from EPFL, Switzerland. A strong advocate for open science, he has released widely adopted resources for the field, including datasets and python packages for whole-slide image analysis.
Talk Abstract
Foundation models have reshaped computational pathology, moving the field from task-specific classifiers to general-purpose encoders trained on millions of whole-slide images. This talk traces that trajectory through the systems my collaborators and I have built: UNI, a patch-level encoder for histopathology, and Threads, a slide-level model linking whole-slide images with molecular profiles; Trident, an open toolkit that broadens community access to these models for large-scale slide processing; and ToxScribe, a vision-language model extending this work to non-human pathology for preclinical drug safety. I will then turn to the agentic step, introducing SlideSeek, a system that reasons over slides, retrieves molecular and clinical context, and orchestrates these components for diagnostic workflows.
We are looking forward to the talk!