Recasting Brain Tissue Simulations as a Machine Learning Problem?

Session Chair

Event TypeMinisymposium

CS and Math

Chemistry and Materials

Physics

Life Sciences

Engineering

TimeWednesday, 7 July 202111:00 - 13:00 CEST

LocationHenry Dunant

DescriptionWith the recent success of machine learning, interest was sparked in various computational domains to use those methods to recast certain simulation problems as machine learning problems. In neuroscience, simulating biophysically detailed models of neurons and brain tissue models has become an important tool, complementing experiments and theory. However, already simulating a single detailed neuron may require the solving of 10,000 differential equations, which is computationally costly and sparks the desire to explore whether a recasting of the problem is possible. This mini-symposium explores this question through three presentations and a panel discussion: A first presentation will describe how quantum mechanics can be recast as a machine learning problem. Next, we will learn about the possibilities of approximating detailed models of neurons through deep artificial neural networks. The third talk will introduce how physics-informed neural networks can be used to compute forward and inverse problems for partial differential equations. Finally, the panel discussion will explore what we can learn from these approaches for recasting the simulation of brain tissue as machine learning problems.

Presentations

11:00 - 11:30 CEST | Approximating Single Neurons with Deep Convolutional Neural Networks | |

11:30 - 12:00 CEST | Many-Body Quantum Mechanics as a Machine Learning Problem | |

12:00 - 12:30 CEST | Physics Informed Neural Networks for Simulation of PDEs | |

12:30 - 13:00 CEST | Recasting Brain Tissue Simulations as a Machine Learning Problem - Panel Discussion |