Publications
I publish in leading journals including Evolution, Trends in Ecology and Evolution, Genetics, PLoS Computational Biology, Complexity, IEEE Transactions on Evolutionary Computation, Philosophy and Biology, Biology Direct, Robotics and Autonomous Systems, Artificial Life, Journal of Theoretical Biology, Biosystems, Evolutionary Computation, Entropy and Royal Society Interface Focus.
I am author of "Compositional evolution: the impact of sex, symbiosis and modularity on the gradualist framework of evolution" in the Vienna Series in Theoretical Biology, with MIT Press.
Full list of references on Google Scholar.
A selection of papers is presented below...
Unification of Evolution and Learning
Evolution is sometimes likened to a learning process. But does that mean something exciting, i.e. evolution is clever like other learning systems, or does it mean something entirely dull, i.e. trial and error learning systems are as dumb as evolution? To answer this it is useful to separate biological evolution (the natural phenomenon) from any particular theory about how biological evolution works - such as Evolution by Natural Selection (ENS). ENS is dumb - a mindless mechanical process (short-sighted) - equivalent to trial and error learning. This kind of learning is very limited. Learning systems in general are not like this. They are smart - able to use past experience to anticipate future consequences intelligently. So can biological evolution do that? Is it dumb and mindless, like ENS says it is, or is it smart - an intelligent adaptive process that learns from past experience? -- Obviously, the standard answer is that it has to be dumb - because intelligence is a product of evolution, not evolution a product of intelligence. The conventional idea is that intelligence requires specialised machinery (like neurons and brains) and most living things dont have that - so intelligence is rare in the tree of life and late in the evolutionary story. If this were true, learning could not be part of the general mechanisms of evolution. But what if learning came first? Is that possible? Yes, in fact, learning does not need neurons or brains, or any special machinery. It occurs spontaneously in physical systems with suitable properties - even in systems as simple as a network of springs. We call this 'natural induction'. And natural induction can be a leader in the evolutionary process, solving problems by learning and then showing natural selection where to go - hence 'Evolution by Natural Induction'. This means that biological evolution can be much smarter than the ENS process that Darwin described. Biological evolution (in this context) can be formally unified with associative learning mechanisms familiar in connectionist models of cognition and learning (i.e. neural networks).
Main works include:
►Natural Induction (pdf) - learning and adaptation in physical systems. No evolution, no natural selection, no biology - but significant problem-solving ability.
►Evolution by Natural Induction (part 1 and 2) (pdf) - how natural induction and natural selection interact to make biological evolution smarter than ENS.
►How can evolution learn? (pdf) - introducing the learning/evolution analogy and overview of most of the papers below
►The Evolution of Phenotypic Correlations and 'Developmental Memory' (pdf)
►How evolution learns to generalise: Using the principles of learning theory to understand the evolution of developmental organisation (pdf)
►How to fit in: The learning principles of cell differentiation
►How adaptive plasticity evolves when selected against
►Resolving the paradox of evolvability with learning theory: How evolution learns to improve evolvability on rugged fitness landscapes (pdf)
Evolution and Individuality
How do multiple short-sighted, self-interested entities become ‘part of something bigger than themselves’, acting together at a new level of individuality, even when this opposes the immediate individual needs of those parts? The evolutionary transitions in individuality, such as the transition from unicellular life to multicellular organisms, are poorly understood. Several topics of the 'extended evolutionary synthesis', often treated merely as 'add-ons' to the standard model, are actually vital to creating and sustaining Darwinian individuality.
► The Collective Intelligence of Development and Evolution - all intelligence is collective intelligence. A spectrum of ANN architectures helps us understand collective intelligence in other areas of biology via the nature of their interaction networks.
► Design for an Individual: Connectionist Approaches to the Evolutionary Transitions in Individuality - main idea: transition = ecological relationships that implement non-linearly separable functions, applied to control of reproduction
►Are Developmental Plasticity, Niche Construction, and Extended Inheritance Necessary for Evolution by Natural Selection? The Role of Active Phenotypes in the Minimal Criteria for Darwinian Individuality - The topics of the extended evolutionary synthesis are not peripheral to evolution, they are core.
►The concurrent evolution of cooperation and the population structures that support it
►Social niche construction and evolutionary transitions in individuality
Adaptive Networks and their self-organisation
Evolution by natural selection is a theory that describes things and how they change in frequency. In contrast, "evolutionary connectionsim" is a theory of the relationships between things and how their organisation changes. See Liology and Complex Systems (care of Jeremy Lent). My research has studied how the organisation of networks is changed by the behaviour of the components on the network, and reflexively, the behaviour of the components on the network is changed by the organisation of the network (a.k.a. adaptive networks). This applies to systems that are not evolutionary units (do not exhibit heritable variation in reproductive success) including ecosystems and societies...
►Optimization in “self‐modeling” complex adaptive systems - This is a foundational model for a lot of the other work. It shows how associative learning can improve optimisation.
►Transformations in the scale of behavior and the global optimization of constraints in adaptive networks - This uses associative learning to re-scale the way the system moves in problem space - its a way of thinking about the effect of ETIs.
►Modular interdependency in complex dynamical systems
►Global adaptation in networks of selfish components: Emergent associative memory at the system scale (pdf)
►What can ecosystems learn? Expanding evolutionary ecology with learning theory
►The Web as an Adaptive Network: Coevolution of Web Behavior and Web Structure
Evolution and Optimisation
Evolutionary algorithms are computational optimisation methods inspired by natural selection. New ways of thinking about biological evolution inspire new computational methods.
► Deep Optimisation: Multi-scale Evolution by Inducing and Searching in Deep Representations (a,b) - An deep learning neural network can be used to search a problem space more intelligently.
► Transforming Evolutionary Search Into Higher-Level Evolutionary Search by Capturing Problem Structure - The principles show that multi-scale evolution (like deep optimisation above) can solve problems that single-scale evolution cannot.
► Is evolution by natural selection the algorithm of biological evolution?
► Reducing local optima in single-objective problems by multi-objectivization
► Reducing bloat and promoting diversity using multi-objective methods
► Modeling building-block interdependency
► Symbiotic combination as an alternative to sexual recombination in genetic algorithms
► A computational model of symbiotic composition in evolutionary transitions
► A building-block royal road where crossover is provably essential (pdf)
► Analysis of recombinative algorithms on a non-separable building-block problem
► A simple two-module problem to exemplify the benefit of crossover (pdf)
