Building models for complex systems depends on professional experience which is very hard to be generalized. How can we build models for complex systems according to the behaviors of the system in an automatic way? As new techniques like Graph Network, Deep Learning, and other techniques, we can reconstruct the dynamical models from time series data. We have developed our own models on data-driven modelling of complex systems. Our models can not only learn the dynamics and make predictions, but also can reaveal the hidden network structures hehind the system.
Based on a Bernoulli network generator and a Markov dynamics learner, this paper proposes a unified framework for Automated Interaction network and Dynamics Discovery (AIDD) on various network structures and different types of dynamics. The experiments show that AIDD can be applied on large systems with thousands of nodes. AIDD can not only infer the unknown network structure and states for hidden nodes but also can reconstruct the real gene regulatory network based on the noisy, incomplete, and being disturbed data which is closed to real situations. We further propose a new method to test data-driven models by experiments of control. We optimize a controller on the learned model, and then apply it on both the learned and the ground truth models. The results show that both of them behave similarly under the same control law, which means AIDD models have learned the real network dynamics correctly.
In this work, we introduce a new framework, Gumbel Graph Network (GGN), which is a model-free, data-driven deep learning framework to accomplish the reconstruction of both network connections and the dynamics on it. Our model consists of two jointly trained parts: a network generator that generating a discrete network with the Gumbel Softmax technique; and a dynamics learner that utilizing the generated network and one-step trajectory value to predict the states in future steps
When predicting PM2.5 concentrations, it is necessary to consider complex information sources since the concentrations are influenced by various factors within a long period. In this paper, we identify a set of critical domain knowledge for PM2.5 forecasting and develop a novel graph based model, PM2.5-GNN, being capable of capturing long-term dependencies. On a real-world dataset, we validate the effectiveness of the proposed model and examine its abilities of capturing both fine-grained and long-term influences in PM2.5 process. The proposed PM2.5-GNN has also been deployed online to provide free forecasting service.