A Neural Input Optimization Framework for Structure Search in Dynamical Systems
The world consists of many dynamical and chaotic systems whose underlying functions can be highly nonlinear. Traditional engineering and scientific methods emphasize simplifying problems, sometimes to the point that valuable information may be lost. Instead, understanding of these systems may require analysis of the underlying dynamics. In the past, the technique of discovering and analyzing strange attractors has yielded some success. However, much work is still needed. Machine learning (ML)-based models such as deep neural networks which are based on nonlinear functions may provide a great set of techniques to help in the discovery of structure in dynamical systems. In this paper, we propose an ML-based approach to search for structure in dynamical systems using Neural Input Optimization (NIO). We discuss our methodology and present promising results. Quantitative results help to validate our proposed NIO methodology for structure discovery.