Published at: arXiv:1608.00771

Published at: arXiv:1612.04496

Available at: arXiv:2005.13621

http://dx.doi.org/10.1111/j.1529-8817.2012.01176.x

Our novel agent-based technique simulates morphological dynamics in living tissue; this paper addresses diatom morphology and the technique has since been applied to cancer tumour morphology. This work has led to the founding of a morphogenesis modelling laboratory at Harvard. Unlike previous models, we recreate detailed mechanisms and chemical interactions and this paper provides unexpected evidence that a single mechanism produces two different types of morphogenesis. We faithfully reproduce rib-domination phenomena and our model, experimental methodology, and results, were assessed against state of the art techniques and SEM data by the Natural History Museum.

This invited paper in the journal special issue on “Bio-Inspired Learning and Intelligent Systems” presents our Genetic Programming technique for portfolio optimisation; the first such technique to propose preserving diversity of dynamic phenotypic behaviour, and optimisation of investment returns rather than price-prediction. It is robust to market volatility, reducing down-time while gathering data for retraining, and demonstrates a 3-times performance improvement. The technique is being used by SIAM Capital, is included in Hitoshi Iba's recent book on Evolutionary Computation in Financial Engineering, and has been shown to have superior performance to Support Vector Machines.

DOI: 10.1177/0037549709106692

The novel Complex-Event formalism described in this paper has since been applied by oncologists to model cancerous tumour growth, with specific success in the field of colonic crypt tumorigenesis. The journal was ranked 3rd out of 14 journals covering the field of computer simulation. The formalism supports the modelling of phenomena that are distributed simultaneously across different physical and temporal levels; superiority over alternative modelling techniques has been independently validated, and as further validation this paper demonstrates how it extends the established X-Machines formalism.

This paper was nominated for a Best Paper award; is now cited in Hitoshi Iba's key new book surveying the field of Evolutionary Computation in Financial Engineering; was supported by Thomson Reuters; prompted seven invited talks at industry conferences (including Trade Tech (TT) Europe, TT Architecture, TT Liquidity, HiFreq Trade and CSFI); and the work was also presented at two invited talks at subsequent GECCO conferences. Our novel work is the first to identify a significant instability whereby cautious investment strategies can unwittingly become adventurous, and vice versa, following a shift in market regime.

This paper is the first ever to describe a multi-agent simulation method for describing and monitoring emergent behaviour in a way that retains the underlying structure of that behaviour. This supports analysis of the mechanisms at a lower level that give rise to emergent behaviours. Thus, emergent behaviour may be explained by inspecting its causal structure. This method has very wide applicability: in simulations of biology, of financial markets, and of any system containing hierarchies and emergent behaviour. Invited by the program chair to submit an expanded version as a chapter to a new book "Complex Systems In Knowledge-Based Environments".

Winner of a Best Paper award at the top conference in the field. Industry trials commenced. Addresses a serious problem with Genetic Programming (GP) systems operating in a volatile environment, which suffer substantial performance degradation. Analyses prior work on robustness and finds none sufficient to counter the extreme volatility of the financial markets. Identifies extreme segments in training data, and uses a volatility-adjusted fitness metric during segmented training. Rigorous experimentation and statistical analysis on the notoriously volatile Malaysian equities market shows outstanding improvements over standard GP, index tracking, and technical analysis.

Published at the top Genetic Programming (GP) conference. Industry trials commenced. Invited to contribute expanded version to Journal of Soft Computing. Sister paper to "Evolving robust GP solutions for hedge fund stock selection in emerging markets". Addresses a serious problem where GP systems suffer substantial performance degradation in volatile environments. Analyses prior work and finds none sufficient to counter the extreme volatility of the financial markets. Identifies extreme segments in training data; trains separate solutions, which operate as a committee. Explores averaging and voting for robustness. Statistical analysis on the volatile Malaysian equities market shows outstanding improvements over standard GP.

CEC is IEEE's key annual Genetic Programming conference, second only to GECCO. CEC typically has high submission rates (e.g. 660) and 400+ delegates. In 2006 Hornby's Age-Layered Population Structure system was introduced, but with no real-world validation. This paper provides the first ever evaluation of ALPS on a real-world problem - the evolution of non-linear factor models for financial portfolio optimization. We show that ALPS reduces convergence and provides fitter individuals, out-performing both the market index and Standard GP. This is a substantial empirical study with rigorous statistical analysis. A version is being developed for an Investment Fund Manager.

This paper was published in the top conference in the field; it introduces for the first time the important concept of phenotypic behavioural diversity for Genetic Programming, and demonstrates how preservation of the diversity of phenotypic behaviour leads to improved solutions that are more robust and more generic (that is, the achievement of success through learning general relationships rather than curve-fitting). The problem domain is financial portfolio optimisation. The paper describes a specific technique for preserving phenotypic behavioural diversity, and presents the results of a substantial and rigorous empirical investigation using real-world financial data sourced from Reuters. The results have attracted significant industry attention and we are currently working on technology transfer with two investment funds.

(Equation 5 in this paper has an error; a correction is provided in the two papers "Hitchhikers Get Around", and "Royal Road Encodings and Schema Propagation"). Gzipped PDF(6.12M). Gzipped PDF (217k)