The Human Brain Project Will Push the Boundaries of Supercomputing


Bernd MohrWhen the general public hears about the Human Brain Project (HBP), they immediately think about the possible medical breakthroughs the project will enable, like accelerated development of diagnostic tools and treatments for brain diseases or personalized medicine. This is a natural reaction as the project's prime goal is to produce an integrated, multi-level understanding of brain structure and function.

However, if you look deeper, the HBP plan lists twelve objectives of which, surprisingly six are related to information and communication technology (ICT), including:

  • Develop ICT tools to generate high-fidelity digital reconstructions and simulations of the mouse brain, and ultimately the human brain (“Simulate the Brain”)
  • Develop hardware architectures and software systems for visually interactive, multi-scale supercomputing moving towards the exascale (“Develop Interactive Supercomputing”)
  • Develop and operate six specialized platforms dedicated respectively to neuroinformatics, brain simulation, high performance computing, medical informatics, neuromorphic computing, neurorobotics, and a unified portal providing a single point of access to the platforms (“Develop and Operate six ICT Platforms, Making HBP Tools, Methods and Data Available to the Scientific Community”)
  • Develop ICT tools supporting the re-implementation of bottom-up and top-down models of the brain in neuromorphic computing and neurorobotic systems (“Develop Brain-Inspired Computing and Robotics”)
  • Develop ICT tools to federate and cluster anonymized patient data.(“Map Brain Diseases”)
  • Implement a program of trans-disciplinary education to train young scientists to exploit the convergence between ICT and neuroscience, and to create new capabilities for European academia and industry. (“Education and Knowledge Management”)

 

Extreme-Scale Computing: The HBP High Performance Computing Platform Subproject

The HPC Platform Subproject (SP7) is one of the HBP’s 12 operational subprojects. Its mission is to build and manage the hardware and software for the supercomputing and data infrastructure required to run cellular brain model simulations up to the size of a full human brain. SP7 will make this infrastructure available to the consortium and the scientific community worldwide.
Central to the HPC Platform is the HBP Supercomputer, the project’s main production system, to be located at the Juelich Supercomputing Centre. The HBP supercomputer will be built in stages, with an intermediate “pre-exascale” system on the order of 50 petaflops planned for the 2016-18 timeframe. Full brain simulations are expected to require exascale capabilities, which, according to most potential suppliers’ roadmaps, are likely to be available in, approximately 2021-22. As well as providing sufficient computing performance, the HBP supercomputer will also need to support data-intensive interactive supercomputing and large-memory footprints. Besides the HBP’s main production system in Juelich, there will be a software development system at CSCS, Switzerland, a subcellular computing system at BSC, Spain, and a data analytics system at Cineca, Italy.

While exascale supercomputers will become available sooner or later without any HBP intervention, it is unlikely that these future systems will meet unique HBP requirements without additional research and development. This includes topics like tightly integrated visualization, analytics, and simulation capabilities, efficient European-wide data management, dynamic resource management providing co-scheduling of heterogeneous resources, and a significant enlargement of memory capacity based on power-efficient memory technologies. In line with the European Commission’s established HPC strategy, Forschungszentrum Jülich and its HBP SP7 partners will therefore drive R&D for innovative HPC technologies that meet the specific requirements of the HBP.

Big Data Processing: The HBP Data Integration Strategy

The HBP’s ICT platforms will make it possible to federate neuroscience data from all over the world, to integrate the data in unifying models and simulations of the brain, to validate the results against empirical data from biology and medicine, and to make them available to the world scientific community (see Figure 1). The HBP will consume existing data coming from publications, models, experiments, and patient data stored at hospitals and insurances, but will also generate new data resulting from neuroscience simulations and their analysis.

The HBP Data Integration Strategy [F. Schürmann, EPFL]

Figure 1: The HBP Data Integration Strategy [F. Schürmann, EPFL]

The resulting knowledge on the structure and connectivity of the brain will open up new perspectives for the development of “neuromorphic” computing systems incorporating unique characteristics of the brain such as energy-efficiency, fault-tolerance and the ability to learn. The HBP’s models and simulations will also enable researchers to carry out in silico experiments on the human brain that cannot be done in vivo for practical or ethical reasons.

As one of many examples of the data volumes to be handled by the HBP, consider BigBrain, an ultra-high-resolution 3D human brain model derived from experimental data. The volume of a human cerebral cortex is about ~7500 times larger than a mouse cortex, and the amount of white matter is 53,000 times larger in humans than in mice. A recently published data set of the digitized mouse brain with 1-mm resolution has a total amount of uncompressed volume data of 8 terabytes. The creation of a volume with similar spatial resolution for the human brain would result in around ~21,000 terabytes. The interactive exploration of such a data set is beyond the capacities of current computing. Thus, among other methodological problems, data processing becomes a major challenge for any project aiming at the reconstruction of a human brain at cellular resolution.

Current rating: 4.6