AI in Mathematics Group

Paper on Entropy-Sparsified Regression Learning accepted for publication in PNAS

Prof. Dr. Illia Horenko

Prof. Dr. Illia Horenko

The joint paper "On cheap entropy-sparsified regression learning" by Illia Horenko with Edoardo Vecchi, Juraj Kardos, Olaf Schenk, Andreas Wächter, Terence O'Kane, Patrick Gagliardini and Susanne Gerber was published in the renowned Proceedings of the National Academy of Sciences of the United States of Americs (PNAS).

The paper is about a new mathematics-driven type of AI, "entropy-sparsified regression learning". It has been shown that this method can allow for much more cost- and energy-efficient learning than traditional AI methods such as "deep learning". It has also be shown that it can improve predictions for the dominant climate phenomenon El Nino by a factor of two and that the method allows insights into the physics of the interactions between ocean and atmosphere.

Go to top