Hopfield networks and optimization
Tony R. Martinez, Xinchuan Zeng · 2005
The Hopfield network is an important model in the field of artificial neural networks. One of the most important functions for the Hopfield network is to solve optimization problems for various applications in computer science and engineering. This dissertation presents a series of papers that propose several new algorithms to improve the performance of the Hopfield network for solving optimization problems. The proposed algorithms include rescaling the energy function, utilizing a beam-search mechanism, and combing an evidence-based activation function with a controlled relaxation procedure. Simulation results based on a large number of data sets demonstrate that these new algorithms are capable of significantly increasing the percentage of valid solutions and improving the quality of solutions. This dissertation also presents several papers that address some related topics, which include filtering mislabeled data, approximating an ensemble of classifiers, improving the cross-validation method, and applying neural networks for feature weighting.