Uni-Logo

Direktzugriff

    Artificial intelligence beats human experts again

    29.02.2016 10:56
    IIF team now in the final round of the Automatic Machine Learning Challenge (AutoML)

    A collaborative team of the IIF research groups Machine Learning for Automated Algorithm Design (Dr. Frank Hutter) and Machine Learning (Dr. Joschka Boedeker) won the expert phase of the Automatic Machine Learning Challenge (AutoML). This challenge aims at the development of automatic systems for machine learning on datasets of all kinds (e.g., from medical diagnosis, speech recognition, and object recognition). The study of such data sets by human machine learning experts is traditionally a lengthy and manual process. This competition now aims to automate this process and to make good predictions as quickly as possible.

    Following their successes of beating over 150 teams of human experts in the three previous phases (beginner, intermediate, advanced), in this current expert phase, the IIF team used their fully automated methods to win both 1st place in the competition against human experts and 1st place in a fully-automated contest, thus winning $3000 of prize money. For the first time, the team also won one of the datasets using "Auto-Net", an automatic method for constructing high-performance deep neural networks.

    Machine learning, and in particular deep learning, is rapidly gaining importance in today's information society and the demand for experts far exceeds the capacity of university programs. The fact that the artificial intelligence developed by the IIF team has now repeatedly won competitions against such human experts may thus have far-reaching implications.

    The IIF team is partially funded by the Excellence Cluster BrainLinks-BrainTools, in which machine learning is used to decode brain waves of patients -- e.g., in order to detect the onset of epileptic seizures or to control intelligent protheses. A future goal of the team is to automate this step, so that customized machine learning models can automatically be learned for each patient.

    The team recently presented a scientific article about its methods at the world-leading machine learning conference Neural Information Processing Systems. This article can be found here: http://papers.nips.cc/paper/5872-efficient-and-robust-automated-machine-learning.pdf
    Their system is available online and allows even machine learning novices to use it effectively: https://github.com/automl/auto-sklearn


    More information about the AutoML Challenge can be found here: http://automl.chalearn.org/



    Files:
    16-02_AutoML4_Ausschnitt.png (2.6 Mbyte)

    Benutzerspezifische Werkzeuge