2013/den13

Dendritic Computation in Neurons and Engineered Devices

Members: Benjamin Torben-Nielsen, Daniel Neil, Daniel Rasmussen, Bert Shi, Greg Cohen, Garrick Orchard, Hector Jesus Cabrera Villaseor, Jens Burger, Jonathan Tapson, Amir Khosrowshahi, Kayode Sanni, Omid Kavehei, Michael Pfeiffer, Jennifer Hasler, Qian Liu, Daniel Whittet, Ryad Benjamin Benosman, Ralph Etienne-Cummings, Sergio Davies, Patrick Sheridan, Evangelos Stromatias, Suma George, Tara Julia Hamilton, Timmer Horiuchi, Will Constable, Xavier Lagorce, Zafeirios Fountas

Organizers: Klaus Stiefel (UWS, Australia) Jonathan Tapson (UWS, Australia)

Dendrites are the structures of neurons which collect inputs and integrate & transform them before they reach the spike initiation zone in the axon. A great number of passive and active (voltage-dependent) conductances determine the signal processing happening in dendrites. We use two approaches to obtain an optimized model neuron for a chosen function. The approach is similar to the Neural Engineering Framework, but without the profligate use of neurons; it has also been shown to perform well for time-encoded neural signals. In this work we hope to build on previous work using the Neural Engineering Framework at Telluride, in a way that realigns it more closely with underlying neurophysiology.

After obtaining an optimized model neuron, we compare these models to real neurons, in order to determine which of these will likely compute the function initially chosen. In the context of neuromorphic engineering, we can try to understand the principle of the solution found by the optimization algorithm. Then we can implement it in software, possibly in a simpler model, and solve diverse engineering tasks with it.

A list of possible specific topic area projects. Optimizing dendrites for:

  • Outlier detection.
  • Signal detection with strong background noise.
  • Source separation.
  • Interlacing of two computations (such as thresholding of 2 input populations).
  • Switching of attention.
  • Sensor fusion.

Update: 2 July 2013

We have the following project proposals:

Using SKIM to associate visual, place and odometry cues in RATSLAM (championed by Ralph) Modelling the auditory pathway (championed by Kayode) Examining how to build dendrites with memristors (Jens) Using SKIM to recognise or process cochlear / DVS/ Dynamic disc output (Shih-Chii, Tara) Modeling dragonfly visual attention switching (Klaus) Modeling chagnes in barn owl inferior colliculus (Timmer)

Project pages

* Building SKIM dendrites with Memristors

* Modeling the attention neurons in dragonflies

* The Ripple Pond Recognition Network with SKIM

* Sound Localization with Stochastic Cross Correlation

* Word Recognition with Auditory Spikes

* SKIM in Nengo

* SKIM ON FPAA

Attachments