YES X : "Understanding Deep Learning: Generalization, Approximation and Optimization"
Mar 19 - Mar 22
During the last decade, deep learning has drawn increasing attention both in machine learning and statistics because of its superb empirical performance in various fields of application, including speech and image recognition, natural language processing, social network filtering, bioinformatics, drug design and board games (e.g. Alpha go, Alpha zero). This raises important and fundamental questions on why these methods are so successful, and to what extent they can be applied to a wide range of problems.
Although theoretical results from the 1980s and 1990s already describe the statistical behavior of small neural networks if we assume their parameters can be optimized exactly, this situation is far from what happens in practice. Instead, two crucial features of modern applications are that the number of parameters is much larger than the sample size, and that non-convexity fundamentally prevents optimization methods from finding the globally optimal parameters. In fact, it has become clear that the statistical properties of deep learning are inextricably intertwined with how their parameters are being optimized. To explain the behavior of modern deep learning it is therefore necessary to understand the subtle interplay between generalization, approximation and optimization. Developing such an understanding is particularly important if deep learning is to play a role in more sensitive application areas such as medical practice, self-driving cars, air-traffic control, and so on.
The aim of the workshop is to give a balanced representation of the most recent advances on these topics, from theory to applications, and spanning both statistics, optimization and machine learning topics. The workshop targets primarily (but not exclusively) young researchers, in particular PhD students, postdocs and junior early stage researchers. The workshop will take place over 4 days and consists of tutorial courses and invited talks given by world experts in the field. The tutorial courses will each consist of roughly of 3 hours of lectures and the invited talks will be 1 hour. Furthermore, some of the junior participants will be given the opportunity to present their current work during the workshop by giving a short (30 minutes) oral presentation and possibly poster presentations (depending on the number of submissions).
|Paulo de Andrade Serra||TU Eindhoven|
|Rui Pires da Silva Castro||TU Eindhoven|
|Tim van Erven||Leiden University|
|Botond Szabo||Leiden University|
|Peter Bartlett||University of California - Berkeley|
|Jason D. Lee||Marshall School of Business|
|Johannes Schmidt-Hieber||University of Twente|
|Nathan Srebro||Toyota Technological Institute at Chicago and University of Chicago|
|Max Welling||University of Amsterdam|
Sponsors and Financial Support