Liu Yanzhu, Adams Wai Kin Kong, Wai Lee Chan, Sparsity Guided Co-teaching Neural Networks for Dynamics Identification and Prediction, The IEEE Conference on Artificial Intelligence (IEEE CAI).
Abstract:
Although deep learning methods have achieved promising performance on forecasting the future states of dynamical systems, most of them perform as a black box. This paper aims to discover the underlying equations of dynamical system from noisy observations. A dual Recurrent Neural Networks (RNNs) equipped with a sparse identifier is proposed for this problem. The identifier is optimized to determine the fewest terms in possible differential equations representing the dynamics, and the identification loss is used to co-teach the dual RNNs predicting the future dynamics. Experimental results on two benchmarking dynamical systems demonstrate the proposed method’s capabilities on discovering the governing equations and predicting for future states.
License type:
Publisher Copyright
Funding Info:
There was no specific funding for the research done