{"id":889,"date":"2017-12-19T11:21:53","date_gmt":"2017-12-19T11:21:53","guid":{"rendered":"http:\/\/www.mub.eps.manchester.ac.uk\/in-abstract\/?p=889"},"modified":"2018-04-17T15:34:58","modified_gmt":"2018-04-17T14:34:58","slug":"using-stochastic-spiking-neural-networks-on-spinnaker","status":"publish","type":"post","link":"https:\/\/www.mub.eps.manchester.ac.uk\/in-abstract\/using-stochastic-spiking-neural-networks-on-spinnaker\/","title":{"rendered":"Using Stochastic Spiking Neural Networks on SpiNNaker to Solve Constraint Satisfaction Problems"},"content":{"rendered":"

Brain-inspired solvers: Implementing stochastic optimization with networks of spiking neurons<\/strong><\/p>\n

Constraint Satisfaction Problems (CSPs) are recognised among the computer science community as being computationally hard, in the sense that the time taken to solve them on a conventional computer grows very rapidly as the problem size grows. Now, researchers at the University of Manchester have developed a framework for building stochastic spiking neural networks that are able to solve a sub-class of CSPs with a high probability. The networks run on the SpiNNaker machine and are shown to solve the hardest class of Sudoku problems in a few seconds using networks that run in biological real time. Similarly, other stochastic networks solve map colouring and Ising spin problems. Overall stochastic networks are shown to have similar capabilities to those of the D-Wave quantum annealing machine – the only quantum computer commercially available today. The D-Wave machine is much faster than the SpiNNaker networks, but SpiNNaker is more energy-efficient and quite a lot cheaper!<\/p>\n

<\/p>\n