Overview
Optimisation problems are commonplace in many engineering disciplines. From optimising the fuel efficiency of a jet engine, to minimising the cost of shipping goods around the globe. These optimisation problems come almost always with certain constraints, such as needing to ensure that the turbulance of the jet is within an acceptable level, or keeping the travel time of goods below a certain threshold of days. In most real-world settings, both the objective (e.g., fuel efficiency) and the constraints (e.g., turbulance levels) are unknown and need to be learned from data. Bayesian optimisation is the de-facto standard method to tackle the optimisation of unknown objectives. In this project, we want to extend Bayesian optimisation to tackle the problem of unknown constraints. We propose to do this using Lagrange multipliers, a well-known technique used in classic optimisation.
FAQs
- What will I learn in this Project?
You will learn about Bayesian optimisation: a practical probabilistic machine learning approach for optimising unknown objectives. Under the hood, Bayesian optimisation algorithms rely on Gaussian processes. The algorithm will be made part of an existing Bayesian optimisation software library. The student will have to learn how to develop code as part of a bigger system.
- What is the objective of the project?
The main objective is the development of an algorithm that can solve real-world constrained optimisation problems. In this problem definition, both the objective and the constraints are unknown and need to be learned from interacting with the environment. A second objective is the implementation of high-quality code, that is part of a larger Bayesian optimisation library.
- How does this fit into the bigger picture?
This project is part of a wider project called AutoAI. The aim is to build explainable and maintainable machine learning systems. You will have a chance to interact with the wider team and a successful project will form part of the portfolio of AutoAI demonstrations.
