Uni-Logo

Direktzugriff

    Computer scientist Frank Hutter receives an ERC Consolidator Grant

    18.03.2022 10:03
    Deep Learning 2.0: Automated Machine Learning enables more precise algorithms Photo: Klaus Polkowski

    Photo: Klaus Polkowski

    The computer scientist Prof. Dr. Frank Hutter from the Department of Computer Science at the University of Freiburg has been awarded a two-million-euro Consolidator Grant from the European Research Council (ERC) for his research on Deep Learning. Hutter and his team are developing the next generation of Deep Learning methods, which, in addition to increasing accuracy, should also strengthen the trustworthiness of Deep Learning. The Deep Learning 2.0 methods should also be easy to use by non-experts. The ERC Grant is one of the most prestigious awards for European scientists. The researchers will be funded for five years.

    Deep Learning is a subfield of Machine Learning and uses neural networks as well as big data. Deep Learning models are able to learn on their own. To do this, the systems repeatedly link what they have learned with new content. Deep Learning is used wherever large amounts of data are analyzed for patterns and trends, for example in the context of face or speech recognition. However, modern Deep Learning methods can now also deliver good results even for small data sets.

    With the help of the ERC grant, Hutter will implement the project “Deep Learning 2.0: Meta-Learning Qualitatively New Components,” in which he will develop a new generation of Deep Learning methods. “Deep Learning 2.0 algorithms should not only compute more precisely, but also make better decisions, for example in terms of robustness, algorithmic fairness or explainability,” explains Hutter.

    To this end, Hutter and his team are combining their research field Automated machine learning (AutoML) with existing Deep Learning methods. Deep Learning 2.0 algorithms will use AutoML to build upon and automatically improve Deep Learning 1.0 techniques. For this purpose, the research group is developing efficient meta-level learning methods to replace previous hand-crafted Deep Learning methods with novel, automatically adapted methods that are directly aligned with the users’ goals.

    “Meta-learning can optimize the entire Deep Learning method. With our proposal, we aim to lay the foundation for the next generation of deep learning, which promises higher accuracy and more trustworthy results,” says Hutter. “Deep Learning 2.0 will also be able to take into account various constraints, such as hardware needs, carbon footprint or training costs.”

    Hutter is regarded as one of the co-founders of research on Automated Machine Learning (AutoML) worldwide. "Our work on AutoML is primarily about democratizing machine learning and making this key technology applicable to non-experts via open-source software, for example in small and medium-sized enterprises," says Hutter. He already secured an ERC Starting Grant and an ERC Proof of Concept Grant from the European Research Council for his research before, as well as an Emmy Noether Fellowship from the German Research Foundation (DFG).

    Hutter heads the professorship for Machine Learning at the Faculty of Engineering at the University of Freiburg and is head of the Machine Learning Lab, as well as head of the ELLIS unit Freiburg. He studied at the TU Darmstadt and received his PhD from the University of British Columbia in Canada.

    More details on Deep Learning 2.0 can be found in this blog post by Frank. Hutter.

     

    European Research Council press release (ERC)

     

    Downloadable press photo
    Photo: Klaus Polkowski

     

    Contact:
    Prof. Dr. Frank Hutter
    Department of Computer Science
    University of Freiburg
    Tel.: 0761/203-67740
    e-mail: fh(at)cs.uni-freiburg.de
    Twitter: @FrankRHutter

    Franziska Becker
    Office of University and Science Communications
    University of Freiburg
    Tel.: 0761/203-54271
    e-mail: franziska.becker(at)pr.uni-freiburg.de

     

     

     



    Benutzerspezifische Werkzeuge