EU $1.3 Billion ten year human brain simulation project developing next generation neuromorphic chips and neurorobotics platforms

European Union (EU)’s Human Brain Project provided a 32 page report on their progress toward an artificial brain by 2023.

The 10-year-long $1 billion euro (US$1.3 billion) Human Brain Project aims to simulate the entire human brain on supercomputers first, then build a special hardware emulator that will reproduce its functions so accurately that diseases and their cures can be tried out on it. Ultimately, the long-term goal is to build artificial brains that are inexpensive enough to outperform traditional von Neuman supercomputers at a fraction of the cost.

Junction of four HICANNSs (High Input Count Analog Neural Networks) inside a reticle on a wafer containing many HICANN circuits. (Source: University of Heidelberg)

After the first year
* all personal are hired
* laboratories throughout the region engaged
* information and communications (ICT) researchers in 20 countries to effectively collaborate and share data
* in progress projects are
—- reconstruct the brain’s functioning at several different biological scales
—- analysis of clinical data of diseases of the brain
—- development of computing systems inspired by the brain.

* a brain simulation technique originally proven for the neocortex was successfully repurposed for the cerebellum

There will 30 more months of a ramp up phase.

Cognitive Architectures

Circuits linking perception to action, body perception and sense of self: Electrocorticography (ECoG) signals from perceptual and visuo-motor tasks have yielded important insights into the use of specific frequency bands for carrying forward and backward visual signals. A new neurodynamic model for multi-stability of action perceptions has been compared to experimental data. Advances have been made in the anatomical dissection of the parietal operculum, the inferior parietal cortex and adjacent parts of the superior temporal gyrus based on existing brain imaging data on bodily self-consciousne.

Architectures for Decision and Confidence: A new task has revealed behavioral evidence that human subjects can base both primary decisions and confidence judgments on close-to-optimal computations with statistical distributions

Architectures for learning and memory: An understanding of how working memory and real-life episodic consolidation starts in the human hippocampus has been established. The first model of the generation of “spindle-ripple events”, which mediate the transfer of reactivated memory information from hippocampus to extrahippocampal circuitry during sleep, has been developed.

Architectures for learning and memory: An understanding of how working memory and real-life episodic consolidation starts in the human hippocampus has been established. The first model of the generation of “spindle-ripple events”, which mediate the transfer of reactivated memory information from hippocampus to extrahippocampal circuitry during sleep, has been developed.

Architectures for language, syntax, and social brain: New fMRI experiments have been designed, and subjects have been scanned, to isolate the neural correlates of language and mathematical syntax, and interpersonal representations.

Neuroinformatics Platform

The Neuroinformatics Platform: The architecture and user specifications of the Platform have been developed and published as a Deliverable. Development has started and an internal draft is on schedule for release to the HBP Consortium in mid-2015.

ESPINA software: A new version of the ESPINA software was released. This software is used for automated segmentation of data in electron microscopy image stacks.

Data sets assembled: Data on neuronal, glial and synaptic proteomics, neocortical microcircuitry, whole mouse brain, and whole brain tracts have been identified and are being integrated into the first draft of the Neuroinformatics Platform

Data collaborations: Collaboration with the Allen Institute for Brain Science is underway in the areas of standard protocols, quality control and metadata standards and modelling. A memorandum of understanding was signed with the Visible Brainwide Networks Project from the Britton Chance Center for Biomedical Photonics in Wuhan, China for collaboration on whole brain rodent data and feature extraction. In addition, a collaboration has been initiated with the newly funded Australian Research Council Centre of Excellence for Integrative Brain Function based in Melbourne, Australia.

Brain Simulation Platform

The Brain Simulation Platform: The architecture and user specifications of the Platform have been developed and published as a Deliverable. Development has started and an internal draft is on schedule for release to the HBP Consortium in mid-2015.

Brain simulation software: HBP efforts have been aligned with community-driven roadmaps for the NEST, NEURON, and STEPS simulators, ensuring long-term viability and global community participation. In its first year, the HBP has achieved substantial cale-up of NEST on massively parallel computers (Kunkel et al., Frontiers in Neuroinformatics, 2014), a reduction of the memory footprint by a factor of six in NEURON, and advances in the parallelisation of STEPS.

Molecular models: Prototype software and workflows have been developed for automated loading, distribution and specification of reactions between molecules in neurons.

Cellular models: The automated modelling of neuronal firing behaviour developed on neocortical neurons has been tested on the most complex neuron — the cerebellar Purkinje neuron — and also on human pyramidal neurons.

Microcircuits: A first draft cellular and synaptic reconstruction and simulation of the rodent somatosensory microcircuit with 207 different cell-types and 40 million synapses has been completed and released through an online portal.

Mesocircuits (brain regions): Simulation of a part of the rodent neocortex with 3 million morphological detailed neurons
and 400 million dynamic synapses has been achieved. A first draft reconstruction of the cerebellar network of neurons has also been made.

Neuromorphic Platform

Portable neuromorphic systems: For both hardware systems, small-scale portable neuromorphic computing devices have been produced. These devices are USB add-ons for conventional laptops and are now in use by outside groups and students to explore the potential of neuromorphic computing.

Next-generation chips: New prototype chips of the neuromorphic systems have been developed for both hardware systems. The prototypes represent key ingredients for the realisation of the detailed Neuromorphic Platform roadmap that defines the technical specifications of the Platform until 2023

Neurorobotics Platform

The Neurorobotics Platform: The architecture and user specifications of the Platform have been developed and published as a Deliverable. Development of the Platform has started and the first version will be ready for internal release to the HBP Consortium in mid-2015.

Robot and body models: Several mobile robot models have been integrated and tested in our development system. In addition, a first prototype of a virtual mouse body model has been developed.

Environment models: A complete virtual room, including furniture and computer screens, was created and added to the development system.

Closed loop simulation: The first closed-loop simulations have been run on our development system. The simulations include a robot that moves inside the virtual room and is controlled by a simple point neuron network.

If you liked this article, please give it a quick review on ycombinator or StumbleUpon. Thanks

Subscribe on Google News