Exploration and Acceleration of Novel Mathematical Tools for Simulation of Spiking Neural Networks. [Master Thesis – Theme: BrainFrame]

One of the most difficult challenges in Computational Neuroscience is to find adequate mathematical tools to simulate detailed biophysical models of neuronal networks accurately and fast. The computational demands of such simulations are huge and more so when we consider the problem size: thousands to millions of neurons connected via up to billions of synapses. Many parts of these kinds of simulations have been parallelized to take advantage of high-performance computing (HPC) but some are still serial, inhibiting performance. Most of the mathematical tools used are Ordinary Differential Equation (ODE) solvers that simulate different biological processes. In the NeuroComputing Lab (NCL) of the Erasmus Medical Center, Rotterdam, we have been working on next-generation, high-performance simulators of Spiking Neural Networks (SNNs) for a decade.

The subject of this MSc thesis topic is first to explore the use of a new parallel solver, in terms of performance, stability and accuracy (inspired by electrical network analysis) to accurately simulate the transmission of electrical signals throughout the whole body of a neuron cell. As a second step, the parallel solver will be ported to a parallel platform or accelerator (OpenMP, GPU, FPGA, Dataflow Engine) for assessing performance-scalability aspects. This thesis is part of a novel simulator called EDEN, which challenges the existing status-quo of human-brain simulators, and is part of the BrainFrame theme jointly developed by the Erasmus Medical Center and the Delft University of Technology.

Keywords: ODE, parallel solver, simulator, neural network, hardware acceleration.

Prerequisites: C/C++, Python, knowledge of numerical analysis.

Optionally: Linear algebra, System theory, parallel programming or acceleration.

Contact: Christos Strydis

Apply for this position

Allowed Type(s): .pdf, .doc, .docx