Biologically-Inspired Massively Parallel Computation (BIMPC) Home Page
Biologically-Inspired Massively Parallel Computation (BIMPC):
This five year project, which began in March 2013,
is funded
by an EU
Engineering Research Council (ERC) advanced grant
and its aim is to use our SpiNNaker platform to investigate
new computational frameworks for massively parallel
computation, inspired by the structure and function of real brains.
The brain can be viewed as a computing machine, but one that works in a
very different way from the personal computers we know from our
everyday lives. While your desktop PC may have a single very fast
microprocessor at its heart, connected to separate memory and disk
storage devices using complex binary coded signals, a brain consists
of billions of very simple, relatively slow elements called neurons,
each of which is connected to thousands of other neurons with which
they communicate by sending simple messages in the form of voltage
spikes, perhaps tens or hundreds every second, on average.
But, while conventional microprocessors execute programs consisting
of lists of instructions, the massive parallelism of real brains allow
them to perform incredible feats of computation in a fraction of a
second despite the slowness of individual neurons.
They do this by
breaking up complex tasks like vision, speech recognition and
planning into very many simple computations that can be carried
out simultaneously and the results then combined.
Brains are also good at working with data that is incomplete
or distorted, a capability that we often find difficult
to instil in conventional computers in the form of a program.
One crucial characteristic at the
heart of this skill is that brains have evolved to learn from raw data
directly,
extracting what is relevant in the data it is presented.
Modern techniques of data mining attempt to emulate
this capability,
but there is much progress still to make, particularly in trying to
do this in a massively parallel
system.
In BIMPC we use inspiration from brain structure to propose new
theories of massively parallel
computation to run in real time
on our SpiNNaker platform.
Our work covers a range of different areas:
- Neural memory. Investigating how reliable memories can form in networks of strongly interconnected neurons, how to make them resistant to degradation and how their encoding can change over time to assimilate them, forming a coherent database of knowledge
- Symbolic computation: How to represent symbols and symbol structures using populations of neurons
- Problem solving using constraints: Investigating ways of using big networks to search for possible solutions to constrained problems, like Sudoku
- NENGO and SPAUN: Working with a group at the University of Waterloo in Canada, who developed a neural specification language called NENGO and have used it to produce a 2.4 million neuron multi-function reasoning architecture called SPAUN. We are going to run SPAUN in real-time for the first time, on SpiNNaker!
Background
SpiNNaker is a novel computer architecture inspired by the working
of the human
brain whose development has been funded by the UK's
Engineering and Physical
Sciences Research Council, EPSRC.
A SpiNNaker machine is a massively parallel computing platform, targeted towards three main areas of research:
Neuroscience: Understanding how the brain works is a Grand Challenge of 21st century science. We will provide the platform to help neuroscientists to unravel the mystery that is the mind. The largest SpiNNaker machine will be capable of simulating a billion simple neurons, or millions of neurons with complex structure and internal dynamics.all in real time and in a low power system.
Where to go to find out more:
- Learn more about the SpiNNaker Project
- For more detail on the philosophy of the SpiNNaker Architecture
- The heart of the machine is the SpiNNaker chip
- The System Software running on the machine.
- To access tools and software to run on SpiNNaker systems, see our Downloads page
- Our Support page provides white papers, documents and FAQs.
- The Publications page gives details of papers describing SpiNNaker in detail
For further information on Spinnaker development boards or the Spinnaker project contact us at:
simon.davidson@manchester.ac.uk
Our mail address is:
APT Group,
School of Computer Science,
University of Manchester
Oxford Road,
Manchester
M13 9PL
The design and construction of the SpiNNaker machine was funded by EPSRC
and The University of Manchester.
The ongoing support and software development, with provision of internet
access to the machine, is being supported by the EU through the ICT
Flagship Human Brain Project.
Research using the machine is being supported from the European Research
Council under the European Union's Seventh Framework Programme
(FP/2007-2013) ERC Grant Agreement no. 320689 BIMPC -
"Biologically-Inspired Massively-Parallel Computation".
The research has also received support from ARM Ltd, and from Samsung
through their GRO programme.
We are grateful to all these funding bodies and companies for their support.