The Discrete Leaky Integrate-and-fire Neuron Model Applied to Visual Tracking and Pattern Recognition
Risinger, Lon W.
The Discrete Leaky Integrate-and-Fire (DLIF) neuron uses a simple discrete LIF neuron model that is capable of diverse spatio-temporal behavior. We explore the behavior of the DLIF when driven by periodic (coherent) and constant (incoherent) input. Results show that the DLIF is capable of oscillatory behavior, amplitude to phase conversion, holographic paging, and spike coincidence detection. Exploiting temporal aspect of the DLIF neuron, a network of DLIF neurons is constructed which is capable of motion detection, object tracking i.e. pursuit motion, and behavior similar to micro-saccadic behavior exhibited by humans. Additionally, a network of DLIF neurons is constructed which is capable of pattern recognition utilizing an action potential computation technique based on relative firing times of neurons. A novel learning technique is presented which allows selective hebbian learning in time delayed connections in a feedforward network by manipulating the DLIF leak rate during training.
Neural networks, Pattern recognition, Visual tracking
Risinger, L. W. (2004). <i>The discrete leaky integrate-and-fire neuron model applied to visual tracking and pattern recognition</i> (Unpublished thesis). Texas State University-San Marcos, San Marcos, Texas.