Final Report Summary - BIMPC (Biologically-Inspired Massively-Parallel Computation)
The Biologically-Inspired Massively-Parallel Computing ERC project aims to address these two top-level research questions:
• Can our growing understanding of brain function point the way to more efficient parallel, fault-tolerant computing?
• Can the massively-parallel computing resources that modern technology is delivering be used to accelerate our understanding of brain function?
This bidirectional interplay between computer science and brain science addresses fundamental issues in both disciplines, and is deliberately positioned to maximize the potential for break-through progress in either dimension.
In the course of this work we have proposed and tested a range of neural network models, some of which represent the functionality of biological neural systems in brains and some of which represent more abstract functions. An example of the former is a model of the Basal Ganglia, a region of the brain that is believed to play a role in action selection - deciding which of a set of possible courses of action we should pursue. An example of the latter type is a stochastic - that is, noisy - neural network that can find solutions to a class of hard computational problems, such as the popular number game Sudoku (which has the added benefit of making an accessible demonstration of the system at work).
While we are still a very long way from having a complete explanation of how the brain works, the research carried out in the project has added to our knowledge of the brain and the capabilities of the neurons from which brains are built, and it has also helped establish SpiNNaker as an effective computational platform for testing hypotheses of brain function. It has also raised new issues in neuron modelling which SpiNNaker, being software based, will be able to contribute to addressing in the future.
• Can our growing understanding of brain function point the way to more efficient parallel, fault-tolerant computing?
• Can the massively-parallel computing resources that modern technology is delivering be used to accelerate our understanding of brain function?
This bidirectional interplay between computer science and brain science addresses fundamental issues in both disciplines, and is deliberately positioned to maximize the potential for break-through progress in either dimension.
In the course of this work we have proposed and tested a range of neural network models, some of which represent the functionality of biological neural systems in brains and some of which represent more abstract functions. An example of the former is a model of the Basal Ganglia, a region of the brain that is believed to play a role in action selection - deciding which of a set of possible courses of action we should pursue. An example of the latter type is a stochastic - that is, noisy - neural network that can find solutions to a class of hard computational problems, such as the popular number game Sudoku (which has the added benefit of making an accessible demonstration of the system at work).
While we are still a very long way from having a complete explanation of how the brain works, the research carried out in the project has added to our knowledge of the brain and the capabilities of the neurons from which brains are built, and it has also helped establish SpiNNaker as an effective computational platform for testing hypotheses of brain function. It has also raised new issues in neuron modelling which SpiNNaker, being software based, will be able to contribute to addressing in the future.