Tristan Lazard
I build interpretable machine-learning methods that connect biological images, clinical data, and scientific hypotheses.
I am based in Cambridge at Microsoft Research Health Futures. My work focuses on representation learning for computational pathology, model interpretability, and AI-driven biomarker discovery.
Previously, I completed a PhD in Computer Science at Mines Paris – PSL, supervised by Thomas Walter and Etienne Decencière at the CBIO lab.
Recent updates
-
Our new preprint, Sparse concept attribution for histomorphological hypothesis generation from whole-slide classifiers, introduces SCOPE, a method for generating morphological hypotheses at scale by combining pathology-specific vision-language models with sparse concept attribution. Read the announcement on LinkedIn. - I joined Microsoft Research Health Futures in Cambridge, working on interpretability for scientific discovery in biology!
-
Our paper on predicting intrahepatic cholangiocarcinoma transcriptomic classes from routine histology using self-supervised learning is published in JHEP Reports! Beyond validating SSL-based molecular subtyping on both biopsies and surgical specimens, we uncover an intriguing influence of training-set constitution: good molecular alignment appears more beneficial than raw data quantity for learning meaningful representations. -
Won 3rd place in the VisioMel Challenge: Predicting Melanoma Relapse, organized by the Health Data Hub and the French Society of Pathology. Write-up and code.
I successfully defended my PhD thesis!