From single cells to cognition in software and hardware

Members: Aleksandrs Ecins, Ashley Kleinhans, Chris Eliasmith, Daniel B. Fasnacht, Francisco Barranco, Gert Cauwenberghs, Garrick Orchard, John Harris, Janelle Szary, Jonathan Tapson, Mounya Elhilali, Michael Pfeiffer, Ryad Benjamin Benosman, Sergio Davies, Shih-Chii Liu, Timmer Horiuchi, Tobi Delbruck, Troy Lau, Terry Stewart, Andre van Schaik

- Organized by Kwabena Boahen & Chris Eliasmith

Overall Goals

The over all goals are stated here:  http://neuromorphs.net/nm/wiki/2011/ng11

In sum, the purpose of this workshop is to expose students to software and hardware methods for simulating biologically realistic neural circuits. The benefit of combining these methods is that the Nengo/NEF models do not need to be learned, and so can be constructed rapidly and with reasonably predictable behavior. As well, Neurogrid and other hardware (JAER camera/cochlea, SpiNNaker, etc.) that was used can run spiking neurons in real-time. The projects are intended to explore these methods alone and in various combinations, developing a toolbox for running neural and cognitive models on the next generation of neuromorphic hardware.

Overall Achievements

The achievements of the workshop are best demonstrated by the many models that were constructed, whose specific results are linked to below. These projects fall in several classes, including neuromorphic hardware/software integration, neurally realistic cognitive modelling, and biologically realistic models of specific functions. Together the projects combine to demonstrate a highly successful workgroup. Highlights from each of these categories include:

* Hardware/software integration - Project 4 demonstrates a real time interface between spiking neurons encoding an image using a sparse representation, and a hardware camera that represents image motion with spikes. Project 7 did something similar, but with a spiking cochlea for detecting audio/visual co-occurrence. Projects 2 & 3 demonstrate that the NEF methods can be used to organize two very different kinds of neuromorphic hardware to constructing spiking networks that communicate information, compute specified functions, and implement specified dynamical systems. Project 1 interfaced an NEF path-finding algorithm to a Lego Mindstorms robot.

* Cognitive modelling - Project 9 implemented a playable Tic Tac Toe game using cortical, basal ganglia, and thalamic models in a spiking neural network simulation. Project 10 did the same thing for a magic square problem. Both of these are state-of-the-art in modeling cognition with neurons.

* Neural function - Project 5 implemented a spiking neural model of a controlled oscillator that closely mimics spike trains observed during swimming behavior of lamprey, including speeding up/slowing down and turning. Project 6 demonstrated that a standard machine learning model, SVM, can be implemented efficiently in spiking neurons. This method was successfully applied to an auditory localization problem. Project 8 implemented a new method for performing decisions with a dynamical system in a spiking neural network.

Here are links to all of the projects in this workgroup (also accessible from the workgroup webpage):

  1. Head direction, path integration, hippocampus with a robot [Results]
  2. Nengo networks in Neurogrid hardware [Results]
  3. Nengo networks in SpiNNaker hardware [Results]
  4. Early visual areas driven by motion camera [Results]
  5. Lamprey model using Nengo and Mindstorms [Results]
  6. SVMs in neurons [Results]
  7. Auditory visual fusion [Results]
  8. Working memory for decisions [Results]
  9. Tic Tac Toe [Results]
  10. Magic squares [Results]