Our group investigates new algorithmic techniques for machine intelligence. In particular we focus on two key technologies:
- data mining and data analytics,
- search and optimisation algorithms for complex problems
(computationally hard, poorly specified, multi-objective, multi-dimensional, mixed discrete and continuous, etc).
Algorithms for data mining and data analytics
Robust data mining and analysis tools that are able to cope with peta scale data volumes, high dimensionality and data heterogeneity, and produce human-understandable solutions are urgently needed by almost all aspects of science and society. Our group investigates new scalable algorithms that could cope with a rapidly changing landscape in data mining and data analytic where data volume, velocity, variety, veracity and visualisation are paramount. We apply our methods to big data problems in the natural sciences (e.g., bioinformatics, neuroinformatics).
We aim at producing new algorithms that can: 1. auto-tune thus reducing the overhead for new users, 2. have good scalability to provide accurate predictions for large datasets, 3. align the functioning of the mining/analysis algorithms to new computing infrastructures (e.g. GPUs, cloud platforms) and thus seemingly extracting full efficiency from available computational resources, 4. extract from the analysed data human-friendly visualisations and interpretation as to maximise the uptake of the mined conclusions.
Algorithms for search and optimisation
Search and optimisation algorithms tremendous successes have been concomitant with their "melting" into the fabric of modern life. Thus, it usually goes unnoticed that the cornerstone of many daily activities depend critically on successful optimisation algorithms and that their impact is felt across a vast range of domains (e.g. stock availability in supermarkets, optimal handling of luggage and runaways at airports).
Search and optimisation algorithms aim at providing a solution to the ubiquitous problem of how to successfully orchestrate a balanced trade-off between exploring a complex search space and exploiting available (partial) information about the "shape" that solutions to a complex problem might have. These methods are suitable for the solution of complex problems where standard, efficient methods (i.e. exact or approximation algorithms) do not exist.
Our group develops algorithms that use global and local search to explore and exploit the solution space. Exploration is performed by Memetic Algorithms, Particle Swarm Optimisation, Ant Colony Optimisation or Artificial Immune Systems, while exploitation is commonly done through the use of local search and domain specific heuristics. We have applied these methods to problems in combinatorial optimisation, continuous optimisation, dynamic optimisation and multi-objective optimisation.
The Logistics of Small Things
While traditional search and optimisation research is concerned with developing decision support systems for the optimal handling of macroscopic objects (e.g., truck delivery systems, personnel management, space allocation, runway scheduling, examination timetabling), nano and bio sciences, deal with billions of micro/nano-scale objects. Very little research has been done to bridge the gap between optimisation research and nanobiotechnologies for which scale reduction brings an explosion in objects’ quantities and give rise to many logistics bottlenecks. Succinctly put, cutting-edge nanobioscience relies on sophisticated and costly instrumentation, time consuming protocols, expensive consumables and very refined expertise for the successful implementation of experiments that deal with zillions of very small entities (cells, molecules, macromolecules, nanoparticles, etc). This is happening within the realm of systems chemistry, systems and synthetic biology as well as molecular self-assembly and scanning probe imaging.