|Date: Wednesday, April 06, 2022
Location: 2866 East Hall (2:30 PM to 4:00 PM)
Title: A neural network approach to high-dimensional optimal switching problems with jumps in energy markets
Abstract: We consider optimal switching problems represented by a coupled system of forward-backward stochastic differential equations, in which finite-variational jumps in the forward process drive jumps in the backward process representing the value function associated with the switching problem. We subsequently develop a backward-in-time machine learning algorithm that uses a sequence of neural networks and the dynamic programming principle to solve for optimal switching strategies, where the neural network is able to learn to account for the jumps present in the problem. We then apply this algorithm to a variety of problems arising from energy production and scheduling problems, and find that the algorithm performs with accuracy and experiences only minimal slowdowns as dimension increases.
Speaker: April Nellis