Boris Baros

Primarily funded by the National Air Traffic Service (NATS), my DPhil surrounds investigating techniques for optimising air-traffic control in the NATS transatlantic operations, specifically flight-path planning. Such a spatiotemporal stochastic control problem is inherently high-dimensional and entails intractably large state spaces, not to mention the difficulties involved in modelling such environments. Noting then the practical considerations of needing interpretable flight-path instructions and the lack of training time-series data available, traditional methods and results don't seem to apply. 

The beginnings of my research aim to address how we might approach such problems in generality, starting with mean-field analysis of neural network parametrised feedback controls for entropy-regularised empirical loss minimisation. Doing so will provide much-needed theoretical guarantees - arguably important in air traffic control, where the price of error can be unaffordably high.