Functional programming offers new ways of thinking about visualization and graphics. Recent work has for example shown how a non-strict language such as Haskell can be used to reconstruct algorithms for scientific visualization, structuring programs as a fine-grained pipeline operating on a stream of data. This creates new opportunities for tuning performance, and we have already shown that the resulting programs can match, and sometimes outperform, imperative implementations. I am beginning to explore the paradigm of domain-specific languages (DSLs), how these can be used to express visualization tasks, how they support program transformation and compilation, and through this how functional language can provide a powerful vocabulary for visual computing, from application problems down to the stream model of general purpose GPUs.

Research on graphical rendering originally aimed at increasing the 'photo-realism' of images or the speed of rendering. However for many applications of graphics, "non-photorealistic" images are routinely used, because they have qualities that make them appropriate for their intended task. Artists, for example painters and illustrators, have developed a wealth of representation styles, some of which have been influenced by the traditions of other cultures. Ivan Herman and I have set out a rationale for a new approach to depiction, called Minimal Graphics, inspired by Chinese and Japanese painting. A prerequisite is a richer 'semantic' model of the data to be visualized, and to this end I am exploring how concepts from the semantic web can be used to drive the synthesis of representations.

Using mathematical tools originally developed for program specification Phil Barnard Jon May and David Duce and I developed syndetic modelling, a form of usability modelling where the conjoint behaviour of human user and device could formalised and explored using proof. Key to syndetics was ICS, a model of human information processing developed by Phil over many years. A feature of ICS is the integration of affect into cognition, and some progress has been made in using ICS to understand how different styles of representation affect judgement.

With the support of an EPSRC Advanced Research Fellowship, I investigated techniques for large-scale data visualization. Part of this work involved node-link graphs, which arise in diverse applications including molecular models, organisational management and social structures, telecom networks, and biochemical pathways.

Developing meta-data for visualization is important for three reasons:

• we are increasingly concerned with distributed, heterogeneous data (think of applications in bioinformatics, or the idea of "visual analytics") and require tools that can locate and/or combine data;
• in order to visualize that data we may want to use services that someone else (e.g. an HPC center) provides, and need to locate and/or compose multiple services to make applications; and
• to record, compare, share or collaborate with visualization results, we need a way of stating precisely how those results were obtained.

Although meta-data can be provided through low-level ad-hoc annotations, expressed in RDF, a more systematic and maintainable approach is to develop ontologies for visualization; other work on the Semantic Web has lead to promising approaches and tools for building and using ontologies.