Thursday 16 Jan 2020: Weakly supervised machine learning: a REF case study
Dr. Raul Santos-Rodriguez - University of Bristol
Harrison 170 14:30-15:30
In this talk we will consider the problem of training multiclass classifiers with imperfect labelling, in scenarios where data labels consist on subsets of categories that may contain the true class and possibly several noisy classes. We propose a general method to construct loss functions for specific but arbitrary weak labelling problems, that satisfy some desirable properties like properness or classification calibration. Finally, we will present a case study based on the UK's REF assessment exercise.
Raul Santos-Rodriguez is a Senior Lecturer in Data Science and AI with the Department of Engineering Mathematics at the University of Bristol, focused on theoretical machine learning and its applications to several domains including healthcare or music informatics. He was awarded his PhD in Machine Learning in 2011 from Universidad Carlos III, Madrid. After spending some time in industry as a data scientist developing large-scale recommender systems, in 2014 he moved to IPL, Universitat de Valencia as a Research Fellow working under a Juan the la Cierva Fellowship.