Triplet-based STDP on SpiNNaker hardware

Mostafa Rahimi Azghadi, Sergio Davies

Triplet-based Spike-Timing Dependent Plasticity (STDP)

Triplet-based STDP (TSTDP) is a synaptic plasticity rule acting beyond conventional pair-based STDP (PSTDP). This rule not only is capable of showing the learning window of PSTDP, but also it can demonstrate some complicated biological experiments including: (i) those that consider higher order spike trains (e.g. triplets and quadruplets of spikes), and (ii) those that in addition to timing difference between pairs of spikes, bring the rate of spike-pairs into action of changing the synaptic weight. In contrary, the PSTDP rule fails to reproduce these intricate biological experiments. In addition, TSTDP can be faithfully mapped to the rate based Bienenstock-Cooper-Munro (BCM) learning rule. In [VLSI design of the TSTDP rule Download] and [TSTDP shows BCM-like behavior Download], you can find more information on TSTDP and BCM, as well as two different VLSI implementations of TSTDP rule.

SpiNNaker Hardware

SpiNNaker is a novel computer architecture inspired by the working of the human brain (

Implementing triplet STDP for SpiNNaker

In order to transfer the triplet-based learning algorithm to SpiNNaker hardware, the required learning behavior should be implemented as a C program. Furthermore, the required neuron behavior needs to be implemented as a python script and the target network of neurons should be described as a Pynn program. There are also several other C, Python and binary files that contribute to the execution of the described network on SpiNNaker. The main goal of this project is to produce all the required files and programs to transfer triplet STDP algorithm to SpiNNaker.

Experimental setup

SpiNNaker chip can be connected to a computer through Ethernet connection port to receive the required data including the network and learning algorithm descriptions. Then it runs the network and generate outputs describing how spikes arrive and how weight changes. Below picture shows the experimental setup. On the left is the SpiNNaker board which is connected to the lap top computer and runs the described network. On the right is the output of the network, shown as a text file. The user is able to show the spike timings, the synaptic weight value and other required outputs by changing the related parts of the written C code. Also, the network structure, the number of neurons and the experiment running time can be set in the mentioned python file.

Discussion and further work

The next step of this work can be using of the triplet STDP rule in selecting a pattern from a number of patterns with different mean firing rates. As a result, this experiments will demonstrate if our proposed circuit is capable of orientation selectivity, which is an essential feature of neurons in the primary visual cortex.