Research and scholarship
Research and scholarship
Dr Alastair Channon's main research interest is in open-ended evolution, in a vein similar to Tom Ray's work on Tierra (in that the phenotype to fitness mapping is an emergent property of the evolving environment and competition is biotic rather than abiotic) but using neuroevolution in a neutral network-aware paradigm similar to Inman Harvey's SAGA, and with the mid- and long-term aims of overcoming the severe scaling problems exhibited by today's evolutionary algorithms (including the difficulty of formulating evaluation functions for complex behaviours), and evolving true artificial intelligence through natural selection.
He developed and published [2001] the first ever closed artificial system to pass Mark Bedau et al's established statistical “ALife Test” for unbounded evolutionary dynamics. Earth's biosphere (through fossil-record databases) is the only other system to have passed the advanced form of this test. This is a very significant result: potentially a second example of unbounded evolution. The creation of a system capable of passing the test had been identified by Bedau, Snyder and Packard as “among the very highest priorities of the field of artificial life”.
Dr Channon made the test well-grounded even for long-term unbounded evolution in artificial systems, through the first ever method of computing individual genes' adaptive ('normalized') evolutionary activities [2003, 2006]. These refinements have been found by Andrew Stout and Lee Spector [2005] to be crucial in resisting attempts to achieve a classification of unbounded dynamics in “intuitively unlifelike” systems. Dr Channon's work in this area now focuses on using this combination of this evolutionary system and analytical methods to draw generalized conclusions about open-ended evolution that were previously impossible given just the real-world example.
Dr Channon and his PhD students carry out research into the use of evolution to generate increasingly intelligent agents, in a vein similar to Larry Yaeger's Polyworld research but focused increasingly on 3D virtual creatures as first evolved by Karl Sims, to better enable the observation of evolved behaviours as they emerge and (with a view toward open-ended evolution) to provide a more open range of low-level actions. In published work with a past PhD student, Dr Thomas Miconi (now at The Neurosciences Institute), they demonstrated [2005] the first artificial evolution of physically simulated articulated creatures with realistic co-adapted behaviours using general purpose neurons. The previous need for ad hoc (problem-specific) neurons was a barrier to the long-term evolution of general behaviours. Research in this area is now being carried forward by Dr Channon, whose research is focused on the incremental neuroevolution of reactive and deliberative behaviour in articulated virtual creatures [2013, 2015].
Another advance came from Dr Channon and his past students, Tim Ellis and Dr Edward Robinson, with their published [2007] demonstration, for the first time ever, of incremental neuro-evolutionary learning on tasks requiring deliberative behaviours: evolved artificial neural controllers that solve tasks which cannot be solved by reactive mechanisms alone and which would traditionally have their solutions formulated in terms of search-based planning. More recently [2011] Dr Channon published a paper demonstrating that the introduction of both transcription errors (noise in the genotype to phenotype map) and cultural transmission, in the form of learning by imitation, can enable the artificial evolution of behaviours inaccessible to incremental genetic evolution alone.
Through EPSRC project EP/H031936/1 Evolution as an Information Dynamic System (2010-2013), Dr Channon worked with partners Drs Chris Knight, Rok Krasovec, Roman Belavkin and (now) Professor John Aston, from the Universities of Manchester, Middlesex and Warwick (now Cambridge). Our work on mutation rate plasticity established optimal mutation rate control functions, for both artificial landscapes and natural landscapes defined by DNA-transcription factor affinities, and demonstrated general applicability to biology [2014]. Dr Channon used tables of DNA to protein binding affinities to evolve DNA sequences in computer simulations, and a meta-Genetic Algorithm [2011] to evolve fitness-dependent mutation rate mappings that show a close match with the research team's predictions from information theory. He used Nvidia's CUDA many-core (GPGPU) technology to speed up computational experiments from 1.4 years (maximum, per run) to 3 days, enabling a step change in the rate at which the research was able to advance. The team began the follow-on BBSRC project BB/L009579/1 The theory and practice of evolvability: Effects and mechanisms of mutation rate plasticity in Febrauary 2014. In this project we are primarily investigating the effects and mechanisms of density-dependent mutation rate plasticity, although we first extended our previous work to determine optimal mutation rate control under selection [2015].
Through the above Evolution as an Information Dynamic System project, the team also investigated the relationship between mutation rate, genetic loss and population size. Above a critical mutation rate (CMR) fitter genotypes may be outcompeted by those with greater mutational robustness. Expressed in terms of fitness landscapes, narrow high fitness peaks may be lost by a population while broader, lower peaks are maintained; this phenomena is referred to as “survival of the flattest”. Dr Channon and his then-PhD student (now) Dr Elizabeth Aston, together with other members of the EPSRC project team and Dr Charles Day, also from the Evolutionary Systems research group, established that CMR has an exponential dependence on population size in haploid populations [2011]: that as population size falls, the CMR above which fitter alleles are lost transitions unexpectedly from near-constant (the previous assumption in evolutionary biology) to drop exponentially for small populations, leaving them spiralling toward extinction. We subsequently verified that our model closely reproduces the less significant but established mathematical relationship between population size and 'error threshold' (no mathematical model has yet been derived for CMR), and established that the result also holds for diploid populations [2013]. This and the team's related research has lead to our most recently awarded BBSRC project, BB/M021157/1 Adaptive landscapes of antibiotic resistance: population size and 'survival-of-the-flattest', which will start in August 2015. In this the focus is on the small population of a microbe with a newly arisen antibiotic resistance mutation.
Dr Channon leads the Evolutionary Systems research group in the School of Computer Science and Mathematics at Keele University. He is a member of EPSRC’s College of Peer Reviewers, IEEE, the IEEE Computational Intelligence Society, the ACM Special Interest Group for Genetic and Evolutionary Computation and the International Society for Artificial Life.
