Friday 15 July 2016

UCNC day 5

The final (half) day of UCNC in Manchester.

The last invited speaker of the conference was Steve Furber, talking about the SpiNNaker project (SpiNNaker stands for  "Spiking Neural Network Architecture").  After some interesting historical context, he told us of the SpiNNaker machine: one million processors in an asynchronous spiking architecture.  The preliminary machine, with 500,000 cores, was launched 30 Mar 2016, and more cores have been added since.  It can be programmed in the Python PyNN language.  For example, 165 lines of Python are needed for a Sudoku solver, where the neuronal groups inhibit other groups with the same integer value in the the same row, column, or 3x3 cell.  Once a solution has been found, the inhibitory links decrease, and the spiking rate goes up, solving a "diabolical" puzzle in about 10 seconds.  This isn't just a toy: it is representative of complex constraint problems.  So far people have only been running small programs, as they think how to scale up their ideas.  Although each core is a standard processor, exploiting the asynchronous spiking communication requires a different way of thinking.

Then on to the final technical session.  First was a talk on "Model-Based Computation"; an attempt to extend the definition of analogue computation (which implements a model analgous to the problem) in a way that can cover more of unconventional computation.  Then a couple of mathematical talks about chemical reaction system formalisms.  The first, "Towards Quantitative Verification of Reaction Systems" encoded the system in a formal solver to prove properties.  The next, "Reachability Problems for Continuous Chemical Reaction Networks" looked at proving safety properties in systems with continuous values of reactant concentrations.  The final talk was on "Global Network Cooperation Catalysed by a Small Prosocial Migrant Clique", looking at evolutionary game theory in networks with no global knowledge, and how a small clique of cooperators migrating into a network of defectors could change it to a network of cooperators.

So, another conference ends.  Next year, in Arkansas.

After two solid weeks of travel and listening, my brain is full of exciting science, and I need a lot of sleep!  I'm looking forward to getting home for a bit of a rest.

No comments:

Post a Comment