ERC Starting Grant for Dr. Frank Hutter

Direktzugriff

Artikelaktionen

Photo: Klaus Polkowski

  Dr. Frank Hutter will receive an ERC Starting Grant of close to 1.5 million Euros for the next five years. This award is one of the most prestigious ones given by the European union, designed to kickstart the career of promising young scientists. Out of 3,000 applications in this year, 325 projects are funded in 23 European countries. The University of Freiburg received a total of three such ERC Starting Grants (see the full german press release by the University).

About Dr. Hutter's project: „Data-Driven Methods for Modelling and Optimizing the Empirical Performance of Deep Neural Networks“

Deep Learning has recently made headlines for artificial intelligence, for example through its ability to recognize objects in images, understand language, beat the world champion in the Japanese game of GO, or – in Freiburg's excellence cluster BrainLinks-BrainTools – control intelligent robot arms by thoughts alone. "The results of Deep Learning are great, but one of its main problems is its sensitivity", says Frank Hutter. „To make it work you have to adjust dozens of parameters correctly. If just one is set poorly everything can break down." For this reason, researchers often test hundreds of settings to find a good one, but for the large datasets in the age of "Big Data" this blind blackbox approach is too slow. And here, Dr. Hutter's project comes in, to avoid computing times of years per data set: „We're developing intelligent optimization methods that work quite similarly as human experts. They use previous results on similar datasets and autonomously run short experiments on subsets of the data in order to find good settings about 100 times faster.“ With his foundational research, the computer scientiest wants to develop Deep Learning far enough to work at the push of a button, allowing even non-experts to use it effectively. In the University of Freiburg's excellence cluster BrainLinks-BrainTools, Dr. Hutter aims to apply his research to better classify brain signals – for example, to recognize which movement of a robotic arm a subject is thinking about. While good classification models currently take days to find, Dr. Hutter's research aims to solve this task in real time.

Contact:

Dr. Frank Hutter
Department of Computer Science
University of Freiburg
Phone: 0761/203-67740
Email: fh(at)cs.uni-freiburg.de

Fußzeile

Benutzerspezifische Werkzeuge