For external participants: Please join our mailing list to receive announcements of future C4DM seminars: http://c4dm.eecs.qmul.ac.uk/seminars.html
All welcome, no pre-booking required.
Date and Time Wednesday, 27th September 2017, at 4:00pm
Place Room B.R. 3.01, Bancroft Road Teaching Rooms (Building 10), Queen Mary University of London, Mile End Road, London E1 4NS. Information on how to access the school can be found at http://www.qmul.ac.uk/about/howtofindus/mileend/.
Speaker Sebastian Ewert
Title From Structured Dropout to Proximal Methods: Guiding the Learning Process in Meaningful Ways
Video
Slides Slides
Abstract The goal of machine listening is to extract useful information from sound data. The underlying computational methods typically integrate and develop ideas from machine learning, signal processing, statistical modelling and numerical optimization. In this context, data-driven learning processes play a central role, which fit a model to given data. Depending on (statistical) properties of a dataset, however, these processes sometimes do not converge or yield models of insufficient accuracy. In such cases, regularization is an important tool to guide the learning processes in the right direction. General purpose regularizer, however, often correspond to incorrect assumptions about the data distribution, which again leads to errors. In this talk, we will look at a few examples, where new types of regularizers led to a considerable gain in structure, which enabled new applications and a considerable increase in system performance. Further, we will discuss proximal methods and specifically the alternating directions methods of multipliers, which provide an excellent and extensible framework for integrating highly non-differentiable, non-convex and infinity-valued regularizers.
Bio Sebastian Ewert received the M.Sc./Diplom and Ph.D. degrees (summa cum laude) in computer science from the University of Bonn (svd. Max-Planck-Institute for Informatics, Saarbrucken), Germany, in 2007 and 2012, respectively. After a postdoc at the Centre for Digital Music, Queen Mary University of London (United Kingdom), he became lecturer (assistant professor) for signal processing in the centre in 2015. Currently, he is holding a research position in the EPSRC programme Fusing Audio and Semantic Technologies (FAST) and is one of the academic leaders of the QM Machine Listening Lab.