Thesis (M.A.)--The University of Texas at El Paso, 2008
Artificial Neural Networks (ANNs) are parallel computational models comprised of densely interconnected neurons. The application of an ANNs depends on the configuration of the interconnections of its neurons. Traditionally, custom ANNs designs are implemented for solving a specific problem, thus the ANNs configuration remains fixed. Advances in Field Programmable Gate Array (FPGA) routing techniques and Interconnection Networks derived in the use of Reconfigurable Interconnection Networks (RIN). RIN allow the user to change the FPGA interconnection network at runtime, enabling the FPGA to perform different operations with the same hardware resources. This thesis proposes to apply the concept of RIN into ANNs. This enables the ANNs to change its configuration at run time to solve multiple problems that require different configurations. Furthermore, the RIN allows the ANNs to use the paradigm of pipelining and routing techniques to create virtual neurons needed to solve more complex problems
