Brian J. Ross

Member of the Bio-Inspired Computational Intelligence Group. (BICIG)
Curriculum vitae (pdf)
Publications

Education

• BCSc (Hon): University of Manitoba (1984)
• MSc: University of British Columbia (1988)
• PhD: University of Edinburgh (1992)

• AI: computational intelligence, genetic programming, evolutionary computation, multi-objective optimization
• AI applications: biological modeling, evolutionary design

Research statement

My research interests are currently in genetic programming and multi-objective analysis. Initially, I developed some language induction algorithms for algebras with interleaving. Although the algorithms derived had polynomial complexity, the process algebra used as a target language was not too robust. Further research showed that genetic programming is an excellent means for automatically synthesizing process algebraic systems. I first used a CCS-like process algebra as the target language, and developed a Prolog-based GP system for evolving CCS expressions solving a variety of concurrent problems. I am now studying new evolutionary computation techniques for richer concurrent languages. I have been investigating the automatic synthesis of networks encoded in stochastic process algebra, as well as higher-level bio-network modeling languages such as logical gene gates and PIM. The goal is to automatically synthesize bio-networks that could generate given time-course data, for example, changing protein levels over time.

To support this research, I developed a Prolog-based genetic programming system called DCTG-GP (Definite Clause Translation Grammar for Genetic Programming). DCTG-GP lets the user define their target language using a logical context--free attribute grammar. This environment permits the languages grammar and semantics to be unified together. Syntactic and semantic constraints can also be conveniently encoded.

I am also active in research in evolutionary design. The Gentropy system synthesizes 2D textures that match various feature characteristics of one or more target images - all without human supervision. Suites of image analysis tests rank the suitability of candidate textures. Different implementations of the system have used multiple populations and multi-objective search. The latest system incorporates a mathematical model of aesthetics, with the goal of evolving visually pleasing images. One project investigated the evolution of image filters, which attempt to duplicate a target colour palette, while adhering to the aesthetic model. Other research has used genetic programming to evolve procedural textures for 3D surfaces. Currently, we are exploring the evolutionary design of 3D models. This work has investigated automatic synthesis of building architectures, floor plan designs, and generalized 3D model generation using aesthetic models (see example results below).

Evolutionary Design Gallery

Aesthetic 3D Model Evolution (Steve Bergen)	Evolution of Floor Plans (Robert Flack)	Evolving Conceptual Building Architectures (Corrado Coia)	JNetic Textures (Steven Bergen)
Evolving 2D vectorized images (Steven Bergen)	Evolving 2D image filters (Craig Neufeld, Bill Ralph)	Evolving 2D textures using a model of aesthetics (Bill Ralph, Hai Zong)	Evolving 3D procedural textures (Adam Hewgill)

Teaching