Fri, 02 Nov 2018

14:00 - 15:00
L3

Facial phenotyping and biases

Dr Christoffer Nellåker
(Nuffield Department of Women’s & Reproductive Health University of Oxford)
Abstract

Computer vision approaches have made huge advances with deep learning research. These algorithms can be employed as a basis for phenotyping of biological traits from imaging modalities. This can be employed, for example, in the context of facial photographs of rare diseases as a means of aiding diagnostic pathways, or as means to large scale phenotyping in histological imaging. With any data set, inherent biases and problems in the data available for training can have a detrimental impact on your models. I will describe some examples of such data set problems and outline how to build models that are not confounded – despite biases in the training data. 

Subscribe to Nuffield Department of Women’s & Reproductive Health University of Oxford