Risk-averse decision strategies for influence diagrams using rooted junction trees

This paper presents how a mixed-integer programming (MIP) formulation for influence diagrams that is based on their gradual rooted junction tree representation can be extended to incorporate more general modelling features, such as risk considerations and problem-specific constraints. We propose two algorithms that enable our reformulations by performing targeted modifications either to the underlying influence diagram or to the associated gradual rooted junction tree representation. We present computational experiments highlighting the superior computational performance of our reformulation against an alternative state-of-the-art MIP formulation for influence diagrams that, by default, can accommodate those modelling features.