Speaker: Dr. Marius Lindauer, Universität Freiburg

Title: Automated Deep Learning

Abstract:

In my talk, I will cover the following points (and a bit more):

Machine Learning is omnipresence nowadays, for example face and speech recognition on your smartphone and other smart devices. The recent breakthroughs in many modern AI applications are due to deep learning (specifically deep neural networks). That is because deep learning scales better to big data, enables native end-to-end learning, and has established new state-of-the-art performance on many applications. However to get an accurate deep learning model, users have to make many important design choices (e.g., the architecture of a deep neural network). One way to address this problem is the use of recent advances in automated algorithm design (AAD), which can automatically determine an effective deep learning model for a given data set. Using these AAD methods, applying deep learning for new applications gets feasible for non-expert users (e.g., users without a PhD in deep learning). On the other side, general AAD can also be improved through deep learning, which has many further applications, e.g., improving the performance of software for hard combinatorial problems, such as hardware verification, robotics or logistics.

Profile:

Marius Lindauer is a postdoctoral research fellow in the Machine Learning for Automated Algorithm Design group (www.ml4aad.org) at the University of Freiburg (Germany). He works on algorithm configuration and selection using cutting-edge techniques from machine learning and optimization to automatically improve the performance of arbitrary algorithms, including SAT and MIP solvers, but also machine learning algorithms. Marius Lindauer received his PhD at the University of Potsdam, working in the renowned Potassco Group, and is a co-founder of the international research group on COnfiguration and SElection of ALgorithms (www.coseal.net).