QLS Seminar Series - Shuang Gao
Controlling Transmission Neural Networks
Shuang Gao,聽Polytechnique Montr茅al
Tuesday April 7, 12-1pm
Zoom Link:听
In Person: 550 Sherbrooke, Room 189
Abstract:听Transmission Neural Networks (TransNNs), introduced by Gao and Caines (2022), establish a connection between virus spreading models over networks and neural networks. This talk will provide an introduction to TransNNs and present the approximation technique in connection with the corresponding Markovian Susceptible-Infected-Susceptible (SIS) model with 2^n states, where n is the number of nodes in the network. Under technical assumptions, the conditional probability of infection in the Markovian 2^n-state SIS epidemic model is obtained explicitly, enabling the control via dynamic programming within the framework of Markov Decision Processes (MDP). An approximate receding horizon control method is proposed, which offers significant computational savings compared to the dynamic programming solution to MDP with 2^n states, while also yielding less conservative control actions compared to TransNN-based optimal control. Finally, numerical comparisons will be presented among three approaches: (a) dynamic programming solutions for the MDP model, (b) TransNN-based approximate optimal control, and (c) TransNN-based receding horizon control.聽This is joint work with Peter E. Caines.