NEAL: October 18 to March 19

Alberto Vergani

• Dr Vergani started working on the project in January.
• He's managed to write PyNN code in NEST, and has helped writing tests and tutorials for the four components.
• He's in the process of rounding out working knowledge of the platforms, SpiNNaker, and the NRP.
• In the next few months he's going to concentrate on the next milestone (getting the components running on BrainScales), using the components to build agents, and associative memory.

Four Components Milestone Completed

• The four components are parsing, associative memory, cognitive mapping, and planning.
• All run on Nest and two versions of spinnaker.
• You can get them from www.cwa.mdx.ac.uk/NEAL/NEAL.html
• They all have tests to assure things run correctly.
• You can specify the underlying structure (e.g. sentences to parse, basic semantic net, size of the map, and the plan).
• All use the finite state automata pattern. The cogmap and planning use the timer pattern. These two patterns are also available.
• There are tutorials for the parser, associative memory and FSA. We're working on tutorials for the others.
• We're trying to get these running on the collab.
• Problems with STDP on the new SpiNNaker, but does work.

Publications

• CABots and Other Neural Agents, Frontiers in Neurorobotics
• A Spiking Half-Cognitive Model for Classification, Connection Science
• Implementing Rules in Neurons, SGAI 18.
• A Brain-inspired Cognitive System that Mimics the Dynamics of Human Thought, SGAI 18.
• Two Simple NeuroCognitive Associative Memory Models, ICCM
• Neural Rule Based Systems, Cognitive Neuroscience later in March

Other Work Done

• Developed a rule based system component, with Dainius Kreivenas. Not ready for export yet.
• Got an agent running with SpiNNaker on the NRP.
• Got a large associative memory running on the big SpiNNaker machine. This drove some changes to the component mechanism.
• Progress on a proto neuro cognitive architecture combining the rule based systems and associative memory.
• Robot forward model running in NEST. Problems with SpiNNaker dropping packets. (Not HBP.)

Future Work

• Milestone at the end of June is those components running on BrainScales.
• This should work once we've got the state machine running correctly. We'll have to deal with a variance of processing not seen in the simulators.
• Might have to skip cognitive mapping as STDP could be a problem. If so, we'll substitute in the rule based system.
• Final milestone is March 2020. This is two more components on SpiNNaker and NEST. We'll use the rules; we may use text generation but vision, though tricky, is also an option.