Quantitative Analysis and Decision Science research ranges from optimisation, performance engineering and knowledge discovery to quantitative verification and security. The work of the section has led to five best paper awards at major international conferences (Middleware, QEST, SPECTS, ICPE) .
Significant achievements of the Section include novel fluid and compositional analysis techniques with application to performance, storage and energy modelling. Their open source software (e.g. PIPE2 and JMT), has been downloaded more than 100,000 times over the past decade. In computational optimisation the section has pioneered research in decision making in engineering, energy and finance; stochastic and robust optimisation; decision rules; robust Markov processes; optimisation algorithms for big data. The section’s research in discovery sciences has led to the development of the tranSMART system for translational informatics that is now becoming the standard open source platform for medical research and pharmaceutical industry. The success of this led to the €24M eTRIKS European project (Guo, PI) involving the participation of 12 major pharmaceutical companies and medical research institutions.
Impact and Applications:
Optimisation has many interesting uses in real life. A few examples in the case of our research are: airline planning and scheduling; finance and energy. Click on images below to see examples.
|Out of this section a spin-off company was founded, Decisal, which provides airline planning and scheduling optimization. Current aviation users report profit increase of USD 300 per flight, translating to a million per month for airlines with 20-50 aircraft. This profit increase is based on Unified Optimization of airline scheduling, a technique invented at Imperial College. With Unified Optimization models for Fleet, Network, and Crew Planning, are joined to achieve an overall optimal solution. This is an improvement over the sequential optimization method that airlines have been using until recently, where these models were considered independently of each other, despite their strong interaction.|