Research Projects

The limited computational resources in battery-operated devices like hearing aids limit the application of advanced neural network-based interventions. Recent advancements in deep neural networks have led to the development of low complexity neural networks for applications with limited computational resources. These networks, such as the functional link artificial neural network (FLANN), adaptive Volterra network, and Legendre neural network, have a single layer of weight update and are primarily developed on Cover's theorem principles. This project aims to develop and analyze new low complexity single layer neural networks, aiming to improve performance without significantly increasing computational complexity. This may require identifying new basis functions that could form the building block of these networks. Methods for improving convergence behavior will be developed, followed by detailed steady state and transient analysis of the adaptive learning schemes. Convergence analysis of recently developed low complexity neural networks will be performed to provide insights into their performance. The project also aims to reduce computational complexity by using approximate computing techniques, offering similar or slightly deteriorated algorithm performance with a substantial reduction in computational complexity. A convex combination of adaptive networks and neural networks will be performed to mathematically understand their behavior and offer improvement in performance. The newly developed neural networks will be applied for nonlinear audio signal processing to provide effective solutions even in the presence of nonlinearities in the system.