Dr. Hugh Cartwright: Research interests

General Applications of Artificial Intelligence in Science

For a general introduction to the subject, see Hugh M. Cartwright. Using Artificial Intelligence in Chemistry and Biology: A Practical Guide. Taylor & Francis, 2008. Book preface

1. Genetic Algorithms
For an introduction, see: Hugh M Cartwright Introduction to Evolutionary Computation and Evolutionary Algorithms, in Application of Evolutionary Computation to Chemistry, Roy Johnston (ed.), Springer-Verlag, Heidelberg (2004). Recent work includes:


• Studies of the movement and dispersal of atmospheric pollution (the "chemical mass balance problem")
• Optimizing organic synthetic routes
• QSAR studies of anti-AIDS drugs
• Chemical flow shop and scheduling problems
• Industrial batch process planning for waste minimisation

In recent years there have been developments in Swarm Intelligence, which has some links to Genetic Algorithms. In this method, a number of semi-independent "agents" co-operate to find the solution to a problem. The screen shot below shows many such agents (in yellow) moving around in space (their tracks are shown as "tails"), seeking solutions (shown in green). A cluster of agents has flocked around the best solution to the problem, which lies near the top lefthand corner of the simulation box. The number of agents there is in fact so great that the solution itself is not visible on the screen shot.







2. Self-organizing maps in chemistry
Self-organizing maps (SOMs) have a wide range of applicability in assessing data for which analytical relationships are unknown. We have applied SOMs to several areas, such as:
• Analysis of the oxidation of wines
• Analysis of NMR data
• Prediction of the hydrocarbon potential of coalfields (in a combination with Genetic Algorithms)
• Prediction of the biodegradability of polychlorinated biphenyls
• Intelligent tools to predict chemical toxicity

The figure below shows a stage in the development of a large-scale map.


3. Growing Cell Structure Networks
These are applied at present mainly to biochemical data; a typical application is shown in Jason WH Wong and Hugh M Cartwright Deterministic projection by growing cell structure networks for visualization of high-dimensionality datasets. J. Biomed. Inform.; 38(2005) 322-330. doi:10.1016/j.jbi.2005.02.002.


4. Classifier Systems
Preliminary work on classifier systems includes an investigation of their use in controlling the pH of industrial waste.


5. Neural Networks in Chemistry
Recent work in this area includes:

• Rapid on-line detection of atmospheric pollutants in the workplace
• Artificial Intelligence in the prediction of the weathering of oil spills, and
• Hugh M Cartwright and Christopher Jones Neural Network Based Assessment of Materials for Chemical Protection Gloves. Proceedings of EUROCON 2003 (Computer as a Tool), Ljubljana, Slovenia, September 2003, Vol II, pp 319-323.

Neural Networks are learning tools. The figure below shows how the learning error (the mismatch between the network's interpretation of a data set and the correct interpretation) falls steadily during training to a value indicating that the network recognizes all but about 2% of the samples shown to it after about 400 training cycles. A fully-trained network can then be used to recognize and classify samples it has never seen before.





6. Artificial Intelligence in Medical Diagnosis

Work conducted by Jason Wong with colleagues in Dublin includes the analysis of medical data using a range of statistical and AI methods. For example, Jason WH Wong, Gerard Cagney and Hugh M Cartwright SpecAlign - processing and alignment of mass spectra data sets. Bioinformatics, doi:10.1093/bioinformatics/bti300.

7. Other AI areas
Work includes
• Combined Fuzzy-Logic - Genetic Algorithm studies of resin plant dynamics
• Cellular automata study of environmental remediation.




The group is also a member of CoLoS, in which work is being pursued in a variety of areas. Current work is concentrated on

• The development of web-based experiments in science, which we are integrating into the course in Practical Physical Chemistry at Oxford.
• On-line applets to illustrate Nuclear Magnetic Resonance Spectroscopy
• On-line applets to illustrate IR and Laser Raman spectra

In much of our work we use EJS (Easy Java Simulations), which provides a simple and attractive interface to Java applets. An example of output from EJS is given below:







Publications

Some representative papers are listed here.

Hugh Cartwright's home page
Physical and Theoretical Chemistry Laboratory
Oxford University Chemistry