My name is Mario Geiger. I'm working on neural networks. I live in Lausanne CH. I speak French and English. I like to study the dynamics of neural network and develop equivariant architectures. My favorite ice cream flavor is pistachio.

CV

• Research Scientist at NVIDIA (2023-)
• Postdoc with Prof. Tess Smidt at MIT (2022-2023)
• PhD with Prof. Matthieu Wyart [pcsl] (2017-2021)
• EPFL in physics (2012-2017)
• CFC of laboratory worker in physics (2006-2010)

Publications

• [21] Symphony [arXiv]
• [20] Ophiuchus [arXiv]
• [19] A general framework for equivariant neural networks on reductive Lie groups [NeurIPS 2023]
• [18] Phonon predictions with E(3)-equivariant graph neural networks [AI4Mat 2023 (NeurIPS)]
• [17] Dissecting the Effects of SGD Noise in Distinct Regimes of Deep Learning [ICML 2023]
• [16] e3nn: Euclidean Neural Networks [arXiv]
• [15] Cracking the Quantum Scaling Limit with Machine Learned Electron Densities [arXiv]
• [14] SE(3)-Equivariant Graph Neural Networks for Data-Efficient and Accurate Interatomic Potentials [arXiv]
• [13] How memory architecture affects performance and learning in simple POMDPs [arXiv]
• [12] SE(3)-equivariant prediction of molecular wavefunctions and electronic densities [arXiv]
• [11] Relative stability toward diffeomorphisms in deep nets indicates performance [arXiv]
• [10] Landscape and training regimes in deep learning [arXiv] [doi]
• [9] Geometric compression of invariant manifolds in neural nets [doi] [arXiv][git]
• [8] Finding symmetry breaking order parameters with Euclidean neural networks [doi]
• [7] Disentangling feature and lazy training in deep neural networks [doi][arXiv][git]
• [6] Asymptotic learning curves of kernel methods [doi] [arXiv]
• [5] Scaling description of generalization with number of parameters in deep learning [doi] [arXiv]
• [4] A jamming transition from under- to over-parametrization affects generalization in deep learning [doi] [arXiv] [git]
• [3] Jamming transition as a paradigm to understand the loss landscape of deep neural networks [doi] [arXiv]
• [2] A General Theory of Equivariant CNNs on Homogeneous Spaces [NeurIPS 2019] [arXiv]
• [1] 3D Steerable CNNs [NeurIPS 2018] [arXiv][git][youtube]
• [0] Spherical CNNs [ICLR 2018][git]

Selected repositories

• github.com/mariogeiger/nequip-jax JAX implementation of NequIP
• github.com/e3nn/e3nn-jax JAX library for equivariant neural networks
• github.com/e3nn/e3nn PyTorch library for equivariant neural networks
• github.com/mariogeiger/jarzynski hard spheres collisions with no time steps
• github.com/agepoly/wish website interface for the matching problem
• github.com/mariogeiger/grid library to save and load numerical experiments
• github.com/mariogeiger/negamax small rust implementation of negamax algorithm
• github.com/mariogeiger/thinfilm implementation of multilayer refraction and transmission

Reports

• pdf Polynomial Envelope Functions for Graph Neural Networks

Contact

geiger mario at gmail dot com