Browsing by Author "Zhang, Xu"

Spiking neural networks have been used to investigate the mechanisms of processing

in 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 Articial

Neural Networks for pattern recognition, classication and function approximation

problems with fewer neurons. In machine learning, SNNs has been shown to be able

to solve pattern and robotic control. For SNNs to be used for such problems, they

must incorporate some mechanism for learning. Current methods to train SNNs use

learning algorithms which adjust the synaptic weights according to an update rule.

In most cases the weights are modied directly. In potential applications such as

driving plasticity in neural culture (in-vitro) and training neuromorphic chips the

directly manipulation of synaptic weights is not possible. Therefore, indirect algo-

rithms, which cause the SNNs to learn based on some biological learning mechanisms

using stimulation of neurons oer signicant advantages over the existing algorithm

for these real world applications.

Indirect algorithms train the neural network by using external stimuli to modulate

the synaptic strengths of a neural network according to synapses intrinsic mechanisms

for plasticity. The training algorithms have been demonstrated in both Integrate and

Fire 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 Supervised

Teaching Signals. These algorithms are used to solve the temporal and spatial input-

output mapping problem using temporal coding. The other type of problem is to

mapping input output ring rates using rate coding.

To test the algorithms, SNNs are used to control both virtual and real world

robots. For the real world robots with SNNs, known and Neurorobots, two types

of robot localization techniques are used: Optitrack, using ceiling mounted cameras

and onboard markers, and embedded cameras. Both small and large SNNs with

biologically realistic neurons are used to drive the neurorobots are modeled with

input coming from Optitrack or the cameras with GPU accelerated SNN simulator.

The results show that the indirect perturbation and indirect stochastic gradient

algorithms can train an SNN to control the robot to nd targets and avoid obstacles

even in the presence of sensor noise. The results also show that indirect training with

supervised training signals algorithm can train a feedforward network with 1000s of

neurons to process and output the correct movement commands to localize a target

using from real time images captured from an embedded camera. Finally. an indirect

version of the Remote Supervised Method (ReSuMe) algorithm was developed using

a more biologically realistic form of Spike-Timing Dependent Plasticity to produce a

specic temporal pattern of spiking from a group of neurons. The indirect algorithms

developed in this thesis may eventually allow the ability to train in vitro and in

vivo biological circuits to perform specic tasks using patterns of electrical or light

stimulation.

Even though Articial Neural Networks have been shown capable of solving many problems such as pattern recognition, classication, function approximation, clinics, robotics, they suers intrinsic limitations, mainly for processing large amounts of data or for fast adaptation to a changing environment. Several characteristics, such as iterative learning algorithms or articially designed neuron model and network architecture, are strongly restrictive compared with biological processing in natural neural networks. Spiking neural networks as the newest generation of neural models can overcome the weaknesses of ANNs. Because of the biologically realistic properties, the electrophysiological recordings of neural circuits can be compared to the outputs of the corresponding spiking neural network simulated on the computer, determining the plausibility of the starting hypothesis. Comparing with ANN, it is known that any function that can be computed by a sigmoidal neural network can also be computed by a small network of spiking neurons. In addition, for processing a large amount of data, SNNs can transmit and receive a large amount of data through the timing of the spikes and remarkably decrease the interactions load between neurons. This makes possible for very ecient parallel implementations.

Many training algorithms have been proposed for SNN training mainly based on the direct update of the synaptic plasticities or weights. However, the weights can not be changed directly and, instead, can be changed by the interactions of pre- and postsynaptic neural activities in many potential applications of adaptive spiking neural networks, including neuroprosthetic devices and CMOS/memristor nanoscale neuromorphic chips. The eciency of the bio-inspired, neuromorphic processing exposes the shortcomings of digital computing. After trained, the simulated neuromorphic model can be applied to speaker recognition, looming detection and temporal pattern matching. The properties of the neuromorphic chip enable it to solve the same problem while using fewer energies comparing with other hardware. The neuromorphic chips need applicable training methods that do not require direct manipulations of the connection strength.

Nowadays, thanks to fast improvements in hardware for neural stimulation and recording technologies, neurons in vivo and vitro can be controlled to re precisely in milliseconds. These improvements enable the study on the link between synaptic level and functional-level plasticity in the brain. However, existing training methods rely on learning rules for manipulating synaptic weights and on detailed knowledge of the network connectivity and synaptic strengths. New training algorithms that do not require the knowledge of the synaptic weights or connections are needed while they cannot require direct manipulations of the synaptic strength.

This thesis presents indirect training methods to train spiking neural networks,

which can both modeling neuromorphic chips and biological neural networks in vivo, via extra stimulus without the knowledge of synaptic strengths and connections. The algorithms are based on the spike timing-dependent plasticity rule by controlling input spike trains. One of the algorithms minimizes the error between the synaptic weight and the optimal weight, by stimulating the input neuron with an adaptive pulse training determined by the gradient of the error function. Another algorithm uses numerical gradient of the output error with respect to the weight change to control the training stimulus, which are injected to the neural network for controlling a virtual insect for navigating and nding target in an unknown terrain. Finally, the newest algorithm uses indirect perturbation of the temporal dierences between the extra stimulus in order to train a large spiking neural network. The trained spiking neural network can control both a unicycle modeled virtual insect and a virtual insect

moving in a tripod gait. The results show that these indirect training algorithms can train SNNs for solving control problems. In the thesis, the trained insect can and its target while avoiding obstacles in an unknown terrain. Future studies will focus on improving the insect's movement to using more complex locomotion model. The training algorithms will also be applied to biological neural networks and CMOS memristors. The trained neural networks will also be used for controlling flying microrobots.