Indirect Training Algorithms for Spiking Neural Networks based on Spiking Timing Dependent Plasticity and Their Applications

Abstract

Spiking neural networks have been used to investigate the mechanisms of processingin biological neural circuits or to propose hypotheses that can be tested in exper-iments. Because of their biological plausibility and event-based information trans-mission, Spiking Neural Networks (SNNs) have suggested as alternatives to ArticialNeural Networks for pattern recognition, classication and function approximationproblems with fewer neurons. In machine learning, SNNs has been shown to be ableto solve pattern and robotic control. For SNNs to be used for such problems, theymust incorporate some mechanism for learning. Current methods to train SNNs uselearning algorithms which adjust the synaptic weights according to an update rule.In most cases the weights are modied directly. In potential applications such asdriving plasticity in neural culture (in-vitro) and training neuromorphic chips thedirectly manipulation of synaptic weights is not possible. Therefore, indirect algo-rithms, which cause the SNNs to learn based on some biological learning mechanismsusing stimulation of neurons oer signicant advantages over the existing algorithmfor these real world applications.Indirect algorithms train the neural network by using external stimuli to modulatethe synaptic strengths of a neural network according to synapses intrinsic mechanismsfor plasticity. The training algorithms have been demonstrated in both Integrate andFire neurons and more biologically realistic neural networks. In this thesis, four indi-rect methods to drive the synaptic weights to its desired value in a network through Spike Time Dependent Plasticity (STDP) are developed: Indirect Perturbation, In-direct Stochastic Gradient, Indirect ReSuMe, and Indirect Training with SupervisedTeaching Signals. These algorithms are used to solve the temporal and spatial input-output mapping problem using temporal coding. The other type of problem is tomapping input output ring rates using rate coding.To test the algorithms, SNNs are used to control both virtual and real worldrobots. For the real world robots with SNNs, known and Neurorobots, two typesof robot localization techniques are used: Optitrack, using ceiling mounted camerasand onboard markers, and embedded cameras. Both small and large SNNs withbiologically realistic neurons are used to drive the neurorobots are modeled withinput coming from Optitrack or the cameras with GPU accelerated SNN simulator.The results show that the indirect perturbation and indirect stochastic gradientalgorithms can train an SNN to control the robot to nd targets and avoid obstacleseven in the presence of sensor noise. The results also show that indirect training withsupervised training signals algorithm can train a feedforward network with 1000s ofneurons to process and output the correct movement commands to localize a targetusing from real time images captured from an embedded camera. Finally. an indirectversion of the Remote Supervised Method (ReSuMe) algorithm was developed usinga more biologically realistic form of Spike-Timing Dependent Plasticity to produce aspecic temporal pattern of spiking from a group of neurons. The indirect algorithmsdeveloped in this thesis may eventually allow the ability to train in vitro and invivo biological circuits to perform specic tasks using patterns of electrical or lightstimulation.