Teams

The SimLab Neuroscience is organised into five teams:

Large Scale Simulation and Neuromorphic Systems

As neuronal simulations become ever more complex, innovations are required in software and hardware. We are key contributors to the neural network simulators NEST and Arbor. Execution performance and maintainability are our primary concerns. Simultaneously, we enhance usability by eliciting user requirements and developing domain specific languages to express neural models, for example the NEST modelling language NESTML.

On the hardware side, we investigate novel neuromorphic systems in the Advanced Computing Architecture project. Modular hardware approaches combining traditional and non-von Neumann architectures promise accelerated simulations of neural networks and new computational paradigms. Examples of such systems investigated by our team include the SpiNNaker and FPGA-based systems for design space exploration and prototyping.

Team members: Guido Trensch (lead), Pooja Babu, Miriam Kempter, Charl Linssen, Dr.-Ing. Georgia Psychou

Research and Software:

• Accelerated neural network simulations
• Arbor - multi-compartment models of neural networks
• NEST
• NESTio
• NESTML
• SpiNNaker Support for the neural simulator NEST
• Validation methodology for network simulations

Machine Learning and Data Analytics for Neuroimaging

The amount of neuroimaging data to be analysed has increased over the years, pushing traditional workflows to their limits. These workflows include classical image processing methods, complex modelling steps such as diffusion tensor reconstruction, and modern machine learning techniques such as deep learning models. We adapt methods originally developed for workstations or small clusters to scale up to HPC systems at the Jülich Supercomputer Centre.

In addition, we develop, in close cooperation with neuroscientists, new data analytics and machine learning methods for neuroimaging optimized for HPC systems.

Team members: Dr. Kai Krajsek (lead), Rajalekshmi Deepu, Andreas Müller, Matteo Westerwinter

Research and Software:

• Data analytics for high-throughput image-based cohort phenotyping
• VIGAN: Visualization-guided Interactive Graph Analysis
• Deep Learning for segmentation of 3D-PLI images
• Deep Learning for brain extraction from MRI scans
• Deep Learning software for supercomputers

• HeAT

Meta-optimization for bio-inspired networks

Our team develops high-throughput software tools and innovative methods for the optimization of neuroscience simulations with particular focus on high-dimensional parameter spaces, biologically inspired meta-learning, and dynamic connectivity generation. Our contributions to simulation frameworks such as NEST and TVB mainly consist of the implementation of methods to study plasticity and the relationships between structure and function in the brain.

We also provide high-level support to the user community in the efficient usage of HPC infrastructure for simulation and data analysis, with special emphasis on research at the whole brain scale. Moreover, we contribute to the design and implementation of standards to describe data and models in neuroscience workflows, improving integration and reproducibility.

Team members: Sandra Diaz (lead), Sinovia Fotiadou-Kotsopoulou, Dr. Thorsten Hater, Dr. Han Lu, Hanna Mohr, Drs. Ir. Michiel van der Vlag

Research and Software:

• High Performance Computing for the Virtual Brain
• Interactive steering and visualization
• Learning to learn
• NESTio
• Structural Plasticity
• Visual Connectivity Generation
• Learning to Learn (L2L) on HPC
  

Multiscale Simulation and Design

The brain exhibits dynamic and structural features at a wide range of scales: from microseconds to decades in the time domain, and from microns to meters in the spatial domain. Increasingly, questions are being posed that span multiple scales. This requires the development of HPC-amenable simulation and analysis software that is able to bridge the different scales and integrate their respective observables.

Our team contributes HPC expertise to brain simulation software at the interfaces between scales: embodied robotics, whole brain, point neuron, and morphologically detailed scales. We design and develop tools for the deployment of complex co-simulation workflows combining multiple simulators into connected systems. Additionally, we provide architecture support for the EBRAINS research infrastructure built by the Human Brain Project.

Team members: Wouter Klijn (lead), Dr. Marissa Diaz Pier, Ekaterina Zossimova

Research and Software:

• Multiscale co-simulation
• Modular Science – co-simulator framework
• Visual Connectivity Generation
• Arbor
• NESTio
• High Performance Computing for the Virtual Brain
• Human Brain Project
• hpc4neuro Python library
• Learning to Learn (L2L) on HPC
  

Coordination, Communication and Project Management

Our team supports the SimLab Neuroscience in all matters related to the coordination, management and dissemination of projects, networks and alliances, in which the SimLab is engaged.

In particular, we are responsible for the management of two major European initiatives at the intersection of neuroscience and high-performance computing: The EBRAINS Computing Services work package of the Human Brain Project (HBP), and the Interactive Computing E-Infrastructure (ICEI) project, which is creating the Fenix Infrastructure for the HBP and other science communities.

Team members: Dr. Boris Orth (lead), Dr. Maren Frings, Steffen Graber, Dr. Olivera Korculanin

Projects, networks and alliances:

• Bernstein Network Computational Neuroscience
• Fenix/ICEI
• Human Brain Project
• JARA-HPC
• Supercomputing and Modelling for the Human Brain