The challenge of reproducibility of AI algorithms: from omics to digital pathology
Lundi 16 septembre 2019 – 12h00-13h00
Cesare Furlanello, Ph.D. (MPBA: Predictive Models for Biomedicine and Environment, FBK – Fondazione Bruno Kessler, Italy)
Host : Gisèle Bonne
Auditorium de l’Institut de Myologie, bâtiment Babinski
Hôpital de la Pitié-Salpêtrière
entrée 82 bd Vincent Auriol
métro Chevaleret
Abstract: Bridging the gap between top-notch deep learning research and effective clinical impact requires accuracy, broad validation, reproducibility and interpretability. In this talk, I present methods and solutions aiming at identifying disease trajectories from integrative data, discuss coping with selection bias by means of computational Data Analysis Plans, and showcase studies on Deep Learning applied to digital pathology.
Short BioSketch: Cesare Furlanello is a data scientist, head of the FBK Unit for Predictive Models in Biomedical and Environmental Data. He is an expert in machine learning and bioinformatics applied to multi-modal health data (omics, bioimaging, medical phenotypes) with Deep Learning. Full professor habilitation in BioEngineering. Bioinformatics collaborator of US FDA/NCTR, RIKEN, and adjoint Faculty of the Wistar Institute in Philadelphia, he is the Past President of the MAQC Society for reproducibility and quality control over massive data.
