< Back to previous page

Project

Hybrid Prediction + Optimisation

Industry and society are increasingly automating processes, which requires solving constrained optimisation problems. This
includes scheduling of tasks in electricity demand-response programs, vehicle routing problems in transportation, job sequencing
in manufacturing, rostering of personnel and more. However, the solutions found by state-of-the-art constraint programming
solvers are often found to not match user expectations. Solutions are regularly critiqued by domain experts as impractical,
frustrating for people involved or creating unfair situations that a human planner would never propose. As a result, the
technology is not accepted or workarounds like manual processing are done which reduces its potential.
The key to overcome this is to integrate techniques from machine learning that can learn about the context of the environment
and user. This can move the problem formulation process from a one-shot model + solve paradigm to a hybrid prediction +
optimisation paradigm. This project investigates fundamental approaches to machine learning and optimisation, based on the
effect it has on the optimisation. The main challenges or those of the efficiency and scalability of such an optimisation-based
learning approach, how to best hybredize and integration the learning into the optimisation and how to be robust to changes in
the environment.
The goal is to develop techniques for smart optimisation that can automatically and robustly learn from and adapt to its
environment.

Date:1 Oct 2020 →  Today
Keywords:machine learning, discrete optimisation, reasoning
Disciplines:Machine learning and decision making, Knowledge representation and reasoning, Neural, evolutionary and fuzzy computation