SKM 2023 – wissenschaftliches Programm
Bereiche | Tage | Auswahl | Suche | Aktualisierungen | Downloads | Hilfe
DY: Fachverband Dynamik und Statistische Physik
DY 11: Focus Session: Physics Meets ML II – Understanding Machine Learning as Complex Interacting Systems (joint session DY/TT)
Montag, 27. März 2023, 15:00–18:30, ZEU 250
Machine-learning has recently entered and is now transforming many fields of science, enabling discoveries in a data-driven manner. As a scientific method, however, ML often lacks one defining feature: Explainability. We here seek discussions with pioneers in understanding, explaining, and improving machine learning methods from the point of view as a physical system of interacting elements. In fact, the history of approaching neuronal networks and problems of inference and learning as a problem of statistical physics has a long history, with a number of important discoveries early on. The close relation between spin glasses and neuronal networks are being currently exploited to address pressing questions, such as the remarkable generalization properties of neuronal networks despite their massive overparameterization and their behavior reminiscent of renormalization group transformations.
Organized by Sabine Andergassen (Tübingen) and Moritz Helias (Jülich)
15:00 | DY 11.1 | Hauptvortrag: The challenge of structured disorder in statistical physics — •Marc Mezard | |
15:30 | DY 11.2 | Hauptvortrag: The emergence of concepts in shallow neural-networks — •Elena Agliari | |
16:00 | DY 11.3 | Hauptvortrag: Adaptive Kernel Approaches to Feature Learning in Deep Neural Networks — •Zohar Ringel | |
16:30 | DY 11.4 | Interpreting black-box ML with the help of physics — •Miriam Klopotek | |
16:45 | 15 min. break | ||
17:00 | DY 11.5 | Hauptvortrag: Analysing the dynamics of message passing algorithms — •Manfred Opper and Burak Cakmak | |
17:30 | DY 11.6 | Hauptvortrag: Deep Learning Theory Beyond the Kernel Limit — •Cengiz Pehlevan | |
18:00 | DY 11.7 | Solving the Bethe–Salpeter equation with exponential convergence — •Markus Wallerberger, Hiroshi Shinaoka, and Anna Kauch | |
18:15 | DY 11.8 | Making machines untangle the parquet equations — •Samuel Badr, Anna Kauch, Hiroshi Shinaoka, Karsten Held, and Markus Wallerberger | |