Modelling, Simulation and Machine Learning

CCFEA projects / modelling in constraints / computational finance and economics

What is the research?

Modelling involves identifying stake holders and describing their relations mathematically or procedurally. This allows us to find the equilibrium of the system.

Equilibrium can be found mathematically in simple models. In more complex models (which includes most of the real-life models), simulation is the only practical alternative to find equilibrium. Simulations can often help to identify undesirable outcomes (such as a crash in a stock market). Simulations would help us develop policies to block undesirable outcomes.

Machine learning is sometimes used to “learn” optimal strategies. Without machine learning, one has to evaluate one policy at a time, which could be extremely laborious.

Projects within CSEE, Essex

• Vernon Smith (Economics Nobel Prize laureate, 2002) pioneered the idea of wind tunnel testing new auction designs. The idea is generalized to building economic wind-tunnels for market testing and strategy design. A number of projects have been developed in our Software Wind-tunnel project

• In the high frequency finance project (supported by Olsen Ltd), we attempt to model trader behaviour in order to understand the market.

• The automated bargaining project shows how Nash equilibrium can be approximated through reinforcement learning.

• In the flexible workforce management project (sponsored by BT), we study different ways to allocate jobs to technicians.

• Some of the above projects demonstrate the power of techniques in constraint satisfaction and optimization, a discipline in which the group is internationally renowned.

Why modelling?

Modelling has been used extensively, e.g. wind-tunnels for aeroplane and car design, war plans, design of electricity markets.

Could we be wrong? Yes, we'll always make mistakes. We inevitably simplify things. We may leave out crucial relations.

“All models are wrong, but some are useful” (George Box 1987).

Modelling is a cost-effective way to assess a situation. Modelling and simulations help answering "what-if" questions, so they are pretty useful for stress testing. Machine learning enables us to look for (learn) policies and business strategies. Modelling enables us to scientifically evaluate such policies and strategies.

The alternative is to go into business or implement a policy unprepared.

“More calculation is better than less, Some calculation is better than none” (translation, The Art of War by Sun Zi 6BC).

Expertise

• Edward Tsang: computational finance, constraint satisfaction, machine learning
• Qingfu Zhang: optimisation, machine learning
• Hani Hagras: fuzzy modelling and reasoning

(PDF / Slides)