17
Dendritic Computation Group Project Review 19 July 2013

Dendritic Computation Group

  • Upload
    kendra

  • View
    46

  • Download
    1

Embed Size (px)

DESCRIPTION

Dendritic Computation Group. Project Review 19 July 2013. Projects. Modelling dragonfly attention switching Dendritic auditory processing Mesgarani and Chang, in silicio The auditory pathway Processing images with spikes Dendritic computation with memristors - PowerPoint PPT Presentation

Citation preview

PowerPoint Presentation

Dendritic Computation GroupProject Review 19 July 2013ProjectsModelling dragonfly attention switchingDendritic auditory processingMesgarani and Chang, in silicioThe auditory pathwayProcessing images with spikesDendritic computation with memristors Computation in RATSLAMImage processingSKIM on SpinnakerDendritic computation on NengoSKIM model on FPAASpike based cross-correlation

Auditory Pathway

3Audio Signal to Spikes

Neuron firing rate limited by spike delayRectified by the volley principle and phase-lockingPoisson spike train generated for each fiber for hair cellPromotes parallelism and simplicity in processing through stochastic computation 4Dendritic computation with memristorsJens Burger, Greg Cohen

Memristors for Alpha FunctionsUse tunable resistance of memristor to control time constants for charging and discharging of capacitorUse memristor under 2 conditionsWith fixed resistancesWith changing resistances caused by exceeding threshold

6ImplementationMatlab code rewritten in C++ and interfaced to NgspiceCompute each synaptic function in Ngspice and return data to C++ codeUse multi-threading to compute synaptic kernels in parallel

7ResultsCan reproduce results by using RC circuits as alpha functionsWorked with identical RC circtuits (resistive) and different RC circuits (memristive)

8A lot of the computational power lies within the mapping between inputs and synaptic kernelsRequirements of synaptic kernels was rather low and impact of different setups on overall performance is hard to evaluateProof-of-Concept successfulFor parameter and setup exploration we need more computational resourcesComments9Dendritic computation with NengoDaniel Rasmussen

FPAA Implementation for the SKIM model Suma George,Georgia Institute of TechnologyAtlanta

Replacing SKIM hidden layer neurons with a dendrite

Spiking patterns for different Input delays

Spiking pattern for different patterns: Dendrite with varying diameterGenerating random weights

SKIM model hidden layer with a single n-compartment dendrite

Spiking pattern for random input weights

Stochastic Electronics: cross-correlation with neuronsTara Julia Hamilton, Jonathan Tapson, and others

Autocorrelation with a single neuronCrossorrelation with two neuronsBlock diagram of chip

Calibration with square wave inputs gives phase delay in histogram i.e. it works!