Task-Restricted Symmetries in Recurrent Weight Space
S. Dräger.
ICML 2026 Workshop on Weight-Space Symmetries: from Foundations to Practical Applications
Simon Dräger
sfdraeger at gmail.com
I study how neural networks compute and develop methods for analyzing complex dynamical systems.
About
I am a researcher at the Computational Neurobiology Laboratory (CNL) at the Salk Institute, led by Prof. Terry Sejnowski.
I obtained my undergraduate degree in Computer Science from University of Düsseldorf in 2022 and an M.S. in Computer Science from UC Davis in 2025.
Research Focus
Recurrent-network dynamics
Task-dependent symmetries and structured weight perturbations that reveal how recurrent networks maintain memory and generate sequences.
Causal analysis of neural dynamics
Nonlinear time-series methods for distinguishing interactions from dynamical similarity in neurophysiological recordings.
Robust multimodal perception
Diffusion-based 3D object detection, LiDAR–camera fusion, and accuracy–speed–memory trade-offs in perception models.
Publications
3DifFusionDet: Diffusion Model for 3D Object Detection with Robust LiDAR-Camera Fusion
X. Xiang, S. Dräger, and J. Zhang.
ICASSP 2026 · Preprint 2023
Causal Dynamic Resonance
C. Lainscsek, P. Salami, S. Dräger, A. R. Bulsara, S. S. Cash, and T. J. Sejnowski.
Accepted · Proceedings of the National Academy of Sciences (PNAS) · Preprint 2025
EffiPerception: A Plug-and-Play Efficiency Enhancement Framework for 2D and 3D Perception Models
X. Xiang, S. Dräger, and J. Zhang.
PRICAI 2025 · Preprint 2024
Evaluating the Impact of Loss Function Variation in Deep Learning for Classification
S. Dräger and J. Dunkelau.
Preprint · arXiv:2210.16003