Workshop presentation during MODeM 2023
Date:
At the Multi-Objective Decision Making (MODeM) workshop, co-located with ECAI 2023 in Kraków, I presented my workshop paper on multi-objective reinforcement learning for infrastructural maintenance planning.
The talk argued that maintenance planning should not be reduced to a single-objective problem through reward shaping. The utility functions asset managers use in practice are non-linear: a plan that holds the probability of collapse under a threshold at moderate cost is not interchangeable with one that achieves lower expected cost while occasionally exceeding it. Linear scalarisation cannot represent that preference, and so cannot recover the policies practitioners actually want.
This work was subsequently developed into MO-DCMAC, published in Neural Computing and Applications.
Links