Time non-locality in neural network training for enhanced learning efficiency

We introduce a novel approach to neural network training that incorporates time non-locality to improve the exploration of high-dimensional parameter spaces. By dynamically adjusting the memory variables, our method steers the network away from local minima and toward global solutions. This technique leverages temporal dynamics and long-range order to efficiently navigate the optimization landscape, resulting in enhanced training efficiency and higher accuracies. Our findings suggest that integrating time non-locality into neural network training offers a powerful avenue for solving complex tasks with long-range dependencies.