AR

Computational physicist

Alexandre René

Aachen, Germany

Connect

About

At university I studied both physics and mathematics, drawn by the ability of mathematics to achieve deep understanding of the natural world. This fascination continues to drive my research today, and let me to pursue graduate studies in computational neuroscience. I was attracted by neuroscience’s potential for mathematization: brains are unimaginably complex biological networks, and simply cataloging their constituents will not suffice to explain how they work. Therefore we need effective models whose interactions we can understand in terms of mathematics. However exactly because these networks are so complex, creating models that are both accurate and understandable is very difficult. To address this, I developed computational tools which blend machine learning with classic dynamical systems theory. The goal is to learn models which are more expressive than those that can be obtained using dynamical systems alone, but which still provide deep insights into the biology, in ways a pure deep learning approach cannot. Such methods however have a propensity to suggest a large number of plausible models, which lead me to questions about how best to perform model selection in the context of data-driven methods. My current research direction has shifted to focussing on the network structure itself. As a postdoc in Michael Schaub’s Network Science group, I study inference methods for structured graph models. In particular we are interested in graphs with higher-order structure, such as simplicial or cell complexes.

Current institution

RWTH Aachen

Research focus

Data-driven dynamic networks

Topics

Model selection
Random graphs
Bayesian inference
Inverse problems
Effective models
Computational neuroscience
Theory of deep learning
Delay differential equations

Main tools

Markov chain Monte Carlo
Automatic differentiation
Stochastic processes
Neural networks
Statistical physics

Publications
Teaching
Software projects
Academic history

Papers

Selecting fitted models under epistemic uncertainty using a stochastic process on quantile functions
Graph Neural Networks Do Not Always Oversmooth
Learning Interacting Theories from Data
Decomposing neural networks as mappings of correlation functions
Inference of a mesoscopic population model from population spike trains
Mean, covariance, and effective dimension of stochastic distributed delay dynamics
Excitation of complex spin dynamics patterns in a quantum-dot electron spin ensemble
All-optical tomography of electron spins in (In,Ga)As quantum dots
Temperature dependence of hole spin coherence in (In,Ga)As quantum dots measured by mode-locking and echo techniques
Herzberg circuit and Berry’s phase in chirality-based coded qubit in a triangular triple quantum dot