1

Achieving Approximate Symmetry Is Exponentially Easier than Exact Symmetry
Enforcing exact symmetry can provide a strong inductive bias in scientific machine learning, but exact symmetry may be more restrictive …
B. Tahmasebi, M. Weber
Preprint PDF Project
Priors in Time: Missing Inductive Biases for Language Model Interpretability
Interpretability methods for language models often seek meaningful concepts in activations while treating features as independent …
E. Singh Lubana*, C. Rager*, S. S. R. Hindupur*, V. Costa, G. Tuckute, O. Patel, S. Krishna Murthy, T. Fel, D. Wurgaft, E. J. Bigelow, J. Lin, D. Ba, M. Wattenberg, F. Viégas, M. Weber, A. Mueller
Preprint PDF Project
Adaptive Symmetry Discovery for Dynamical System Identification
Dynamical system identification aims to recover system parameters from observed trajectories, and in scientific settings those dynamics …
B. Tahmasebi, M. Weber
Preprint Project
Data Augmentation: A Fourier Analysis Perspective
Data augmentation enforces invariances by adding transformed copies of data according to a known symmetry group, but full augmentation …
B. Tahmasebi, M. Weber, S. Jegelka
Preprint Project
Neural Feature Geometry Evolves as Discrete Ricci Flow
Deep neural networks learn feature representations via complex geometric transformations of the input data manifold. Despite the …
M. Hehl, M.-K. von Renesse, M. Weber
PDF Project
Protein Graph Neural Networks for Heterogeneous Cryo-EM Reconstruction
This work presents a geometry-aware approach to heterogeneous single-particle cryo-EM reconstruction that predicts atomic backbone …
J. Krook*, A. Janson*, J. Andén, M. Weber, O. Öktem
Preprint PDF Project
Unitary Convolutions for Message-passing and Positional Encodings on Directed Graphs
Many real-world graphs are directed, but standard graph neural networks are often designed for undirected edges, and simple directed …
L. Fesser, B. T. Kiani, M. Weber
Preprint Project
Balancing Fairness and Accuracy in Graph Learning via Fairness-Constrained Rewiring
Graph rewiring can improve graph neural network accuracy by mitigating topological pathologies such as oversmoothing and oversquashing, …
J. Wang, L. Fesser, M. Weber
Preprint Project
Higher-Order Learning with Graph Neural Networks via Hypergraph Encodings
Higher-order information is crucial for relational learning in many domains where relationships extend beyond pairwise interactions. …
R. Pellegrin*, L. Fesser, M. Weber
PDF Code Project
Beyond Euclidean - Foundation Models Should Embrace Non-Euclidean Geometries
In the era of foundation models and Large Language Models (LLMs), Euclidean space has been the de facto geometric setting for machine …
N. He, J. Liu, B. Zhang, M. Bui, A. Maatouk, M. Yang, I. King, M. Weber, R. Ying
PDF Project