HBP Research Platforms Released

Mikael Djurfeldt, PDC

The Human Brain Project (HBP) is a ten year long project that started in 2013 and aims to create and operate a research infrastructure – based on Information and Communication Technology (ICT) – to help advance brain research in neuroscience, medicine and computing. Since the article in an earlier PDC Newsletter (see no. 2, 2013 ) the HBP has been working on establishing its six ICT platforms, which were launched earlier this year on the 30th of March.

The six platforms are as follows:

  1. the Neuroinformatics Platform, a web platform and application-programming interface (API) that will allow users to register, search for and examine large volumes of neuroscience data,
  2. the Brain Simulation Platform, a suite of software tools and models that will allow users to construct and perform detailed simulations of the brain,
  3. the High Performance Analytics and Computing Platform, a hardware and software infrastructure that will provide the high performance computing, storage, and data processing capabilities to analyse large sets of complex data and to run sophisticated, detailed brain simulations,
  4. the Medical Informatics Platform, comprising innovative software that will allow users to access and analyse real patient data located in healthcare databases to help understand similarities and differences among brain diseases,
  5. the Neuromorphic Computing Platform, consisting of two configurable, complementary neuromorphic computing systems that will be able to emulate the microcircuits of the brain and apply principles that are similar to the way the brain learns and understands to computers, and
  6. the Neurorobotics Platform, a software and hardware platform that will allow users to connect virtual models of the brain to simulated robot bodies and environments.

The Neuroinformatics Platform

The Neuroinformatics Platform serves as the Human Brain Project’s search engine for distributed data, curated data repositories, brain atlases and knowledge about the brain. The Platform consists of APIs for querying, plus a web-based platform and application programming interfaces for building software applications. Users
can search and collate high quality neuroscience data generated within and outside the HBP. Data can be examined by species, contributing laboratory, methodology, brain region, and data type, thereby allowing functionality not currently available elsewhere. The data is predominantly organised into atlases (namely the HBP Strategic Rodent Brain Atlases and HBP Human Brain Atlases) and linked to the KnowledgeSpace – a collaborative community-based encyclopaedia linking brain research concepts to the latest data, models and literature. 

The Brain Simulation Platform

The Brain Simulation Platform provides scientists with powerful tools to reconstruct and simulate “scaffold models” of the brain and brain tissue in a data-driven fashion. Its development is embedded in one of the HBP subprojects on Brain Simulation, where a tight co-design loop between science and engineering ensures the required substantial technical and scientfic innovations come to fruition. As a result, the unique functionality of the Platform allows novel questions to be addressed, which could not be investigated previously.

The High Performance Analytics and Computing Platform

The High Performance Analytics & Computing Platform comprises supercomputing capabilities at Forschungszentrum Jülich (Germany), the Swiss National Supercomputing Centre (CSCS) in Lugano, the Barcelona Supercomputing Center (BSC, Spain), and the Consorzio Interuniversitario del Nord-Est italiano per il Calcolo Automatico (CINECA, Italy) in Bologna. The Platform also provides access to high-fidelity visualisation systems at two locations in Germany (RWTH in Aachen) and Switzerland (EPFL in Geneva) and cloud storage with a dedicated capacity of 3 petabytes at the Karlsruhe Institute of Technology (Germany), which will also contribute research and development work to complement these resources. 

All of these systems are connected logically through UNICORE (Uniform Interface to Computing Resources) and physically through the high-speed PRACE (Partnership for Advanced Computing in Europe) network, with a bandwidth of 10 Gigabits per second, and other secure connections. These high-speed connections mean that the users can run larger and more complex workflows than previously possible, and will increase the speed at which research work can be performed. 

The Medical Informatics Platform

Hospital and other medical databases contain vast amounts of data about health and disease that represent an enormous asset to researchers. Many such databases and repositories are largely underused due to issues of data privacy and patient confidentiality, or because it is not possible for researchers outside these institutions to access the material. 

The Medical Informatics Platform is an innovative data analysis system that provides an interface through which clinicians, neuroscientists, epidemiologists, researchers, health managers and even the general public can access and analyse imaging and clinical data currently locked in hospital and research archives and public databases. The identifying details for the patients in the data have been removed (in other words, the data has been de-identified) and the resulting information has been aggregated before working out the overall results so the privacy of the patients is completely protected. Users can explore ontologies and variables, configure and apply statistical methods on clinical and research data, and visualize and dynamically interact with the results. 

The Neuromorphic Computing Platform

The Neuromorphic Computing Platform consists of two complementary and configurable neuromorphic computing systems, based on custom hardware designs. These systems are designed to emulate neural microcircuits and apply brain-like principles in machine learning and cognitive computing, that is, principles that will allow the machine to learn in the way that the brain does rather than being programmed like a normal computer. To do this, the Neuromorphic Computing Platform uses state-of-the-art electronic component and circuit technologies and incorporates new knowledge gained from other areas of neuroscience research, such as experimental neuroscience, theoretical neuroscience and brain modelling. 

The Neuromorphic Computing Platform provides remote access to large-scale neuromorphic computing systems based in Manchester, UK, and Heidelberg, Germany. Together with the TrueNorth system by IBM, these constitute the only neuromorphic systems in the world capable of running simulations of neural circuit systems with state-of-the-art models of neurons, synapses and plasticity in either real time or accelerated time (which, in this case, is equivalent to 10,000 times faster than real time). 

The Neurorobotics Platform

The Neurorobotics Platform is an internet-accessible simulation system that lets users simulate robots and environments controlled by spiking neural networks. 

The Platform enables simple virtual closed-loop experiments in cognitive neuroscience to be performed using brain models developed within the HBP, with the capability to customise several variables, such as the environmental and physical parameters, using a Robot Designer, an Environment Builder and a Closed Loop Engine.

These six HBP Platforms embody the key objectives of the HBP: to gather and disseminate data describing the brain, to simulate and build models of the brain, to develop brain-inspired computing and robotics, and to create a global scientific community around the developing research infrastructure. The development of the six platforms has been the result of an unprecedented multidisciplinary effort involving more than 750 scientific collaborators and engineers from 112 institutions in 24 countries. 

Swedish Participation in HBP

The Swedish partners in the Human Brain Project are Uppsala University (Kathinka Evers), the Karolinska Institute (Sten Grillner and Kevin Grimes), KTH (Jeanette Hellgren Kotaleski and Erwin Laure) and the Linnaeus University (Abdul Mohammed). Anders Lansner was also part of the KTH participation during the start-up phase. Swedish researchers are participating in various areas of the project such as ethics, neuroinformatics and computational neuroscience.

During the start-up phase, Anders Lansner’s group was part of the neuromorphic subproject of HBP. Their activities included the development of benchmark models, such as a detailed model of layers two and three of the cerebral cortex, which can run on both supercomputers and neuromorphic hardware. The PDC Cray XC30 Milner was extensively used in this work.

Jeanette Hellgren Kotaleski from the Department of Computational Science and Technology at KTH is the deputy leader of the Brain Simulation Platform. Her group is currently using Milner at PDC for computer simulations of the basal ganglia and for the development of cell models for the Brain Simulation Platform (see the article Milner Update  elsewhere in this newsletter).

Mikael Djurfeldt who is based at PDC is a member of the neuromorphic subproject in HBP. He uses Milner for the development of two software tools for large-scale neuronal network simulations, MUSIC, which facilitates co-simulation of neuronal network models (see the Milner Update  article in this newsletter), and CSA, which is a library implementing a mathematical method for describing neuronal network connectivity. CSA is used to facilitate the setup of such connectivity in large-scale models.

PDC has also offered computer time on Milner, as well as access to neuromorphic hardware to the Human Brain Project – an offer which has been appreciated by the project and which may be taken up later.

The HBP Contribution to the Neuroscience Community

The six HBP Platforms will enable new kinds of collaborative research to be performed in neuroscience, medicine and computing. The prototype tools, hardware systems and initial data sets are designed to make faster and more efficient research techniques possible in areas such as modelling, in silico experimentation, or data analysis. Users are encouraged to explore the Platforms and to build interactive “collabs” as part of the HBP Collaboratory, a scientific research hub that is accessible via the web and that serves as the main entry point to the Platforms. 

The HBP Collaboratory and the HBP Platforms are subject to restrictions on their use. In most cases these restrictions are due to limited computing or storage capacity powering the Platform service offerings. For example, supercomputing time and an account with a high performance computing centre are required by the High Performance Analytics and Computing and the Brain Simulation Platforms. The Neurorobotics Platform has a limited number of servers that can be used concurrently, and the initial data sets in the Neuroinformatics and Medical Informatics Platforms are limited. If you are interested in using the HBP Platforms, please see www.humanbrainproject.eu/en/hbp-platforms/getting-access  for more details and guidance about accessing the platforms, as well as the terms of service and conditions of use for each of the platforms.