Prof. Dr. Patrick Cramer (Max-Planck-Gesellschaft)
Abstract:
Despite the availability of vaccines, there remains a need for safe and effective antiviral drugs against coronaviruses to treat vulnerable patients, control transmission, and prepare for future outbreaks. In the spring of 2020, our lab was part of a global race to the 3D structure of the coronavirus SARS-Cov2 RNA-dependent RNA polymerase and provided mechanistic insights into viral RNA replication. We later also used structure-function studies to reveal the mechanisms of antiviral drugs such as remdesivir and molnupiravir, revealing key limitations in their efficacy and safety. Building on this framework, we are identifying novel polymerase inhibitors through large-scale chemical screening, in the hope to uncover leads for the development of next-generation antiviral therapies.
10:00 • ETH Zentrum, Building HIT, Room E 51
Cristopher Moore (Santa Fe Institute)
Computational complexity and phase transitions
10:15 • ETH Zentrum, Rämistrasse 101, Zürich, Building HG, Room G 19.1
Kaibo Hu (University of Oxford)
Finite Element Tensor Calculus
10:15 • ETH Zentrum, Rämistrasse 101, Zürich, Building HG, Room G 43
Yatin Dandi (EPFL)
Deep Learning as Neural Low-Degree Filtering: A Spectral Theory of Hierarchical Feature Learning abstract
Abstract:
Understanding how deep neural networks learn useful internal representations from data remains a central open problem in the theory of deep learning. We introduce Neural Low-Degree Filtering (Neural LoFi), a stylized limit of gradient-based training in which hierarchical feature learning becomes an explicit iterative spectral procedure. In this limit, the dynamics at each layer decouple: given the current representation, the next layer selects directions with maximal accessible low-degree correlation to the label. This yields a tractable surrogate mechanism for deep learning, together with a natural kernel-space interpretation. Neural LoFi provides a mathematically explicit framework for studying multi-layer feature learning beyond the lazy regime. It predicts how representations are selected layer by layer, explains how emergence of concepts arises with given sample complexity, and gives a concrete mechanism by which depth progressively constructs new features from old ones through low-degree compositionality. We complement the theory with mechanistic experiments on fully connected and convolutional architectures, showing that Neural LoFi improves over lazy random-feature baselines, recovers meaningful structured filters, and predicts representations aligned with early gradient-descent feature discovery with real datasets.The talk is based on joint work with Matteo Vilucchio, Luca Arnaboldi, Hugo Tabanelli, and Florent Krzakala (https://arxiv.org/abs/2605.13612).
13:45 • ETH Zentrum, Building TBD, Room
Alan Reid (Rice University)
Arithmetic groups and their profinite completions
14:15 • ETH Zentrum, Rämistrasse 101, Zürich, Building HG, Room G 43
Dr. Jonghwa Park (ETH Zürich)
Title T.B.A.
17:15 • ETH Zentrum, Rämistrasse 101, Zürich, Building HG, Room G 43