The nexus for University College London Evolutionary Algorithms Research

Previous talks:

Date Speaker Title
8/4/2009 Navneet Bhalla, University of Calgary Designing Self-assembling Systems
Self-assembly and developmental processes go hand-in-hand with design, through evolution by means of natural selection, to create the plethora of complex biological systems we see throughout nature. However, designing and creating artificial self-assembling systems continues to be extremely challenging. One aspect that remains an open problem is how to design a set of components and their environmental conditions such that the set of components self-assemble into a desired entity. Techniques have been developed for a few specific applications, but a general approach has not been achieved. The focus of my research is on the design aspects of self-assembly. I will present key results of my research over the past five years: from the creation of a fish-like structure from simple mechanical components of varying form, to the creation of a virtual coffee mug from generic spherical components, and finally to the creation of simple mechanical structures resulting from components following a rule-based approach. These results form the basis to the bottom-up design process I am currently developing, for which I will provide an overview. As well, I will present preliminary research on the incorporation of rapid prototyping to aid in designing self-assembling systems. 
Date Speaker Title
16/2/2006 Dr Daniel Polani, University of Hertfordshire Organization of Information Flows in the Perception-Action Loop of Agents
Information is an essential and omnipresent resource and has long been suspected as a major factor shaping the emergence of intelligence in animals and as a guideline to construct artificial intelligent systems. In search for fundamental principles guiding the self-organization of neural networks, Linsker (1988) formulated a number of information-theoretic hypotheses. His model (and most of its successors) was purely passive. However, recent work by Touchette and Lloyd (2000) extending early work by Ashby (1953), as well as some work by Polani et al. (2001) has shown that actions can be incorporated into the information- theoretical analysis.

As was found by Klyubin et al. (2004), incorporating actions into an information-theoretic formalization of the perception action-loop of agents has dramatic consequences in terms of self-organization capabilities of the processing system. As opposed to Linsker's model which required some significant pre-structuring of its neural network, this new model makes only minimal assumptions about the information processing architecture. The agent's "embodiment", i.e. the coupling of its sensors and actuators to the environment, is sufficient to give rise to structured pattern detectors driven by optimization principles applied to the information flow in the system.

In the present talk, we will motivate Shannon information as a primary resource of information processing, introduce a model which allows to consider agents purely in terms of information and show how this model gives rise to the aforementioned observations. If there is time, the talk will discuss the use of information-theoretic methods to structure the information processing also in real robot systems. 
Date Speaker Title
6/10/2005 Prof. Risto Miikkulainen, University of Texas at Austin Solving sequential decision tasks with Neuroevolution
Intelligent agents in the real world often have to make several decisions before they get feedback on how well they are doing: for example, operating a vehicle, playing a game, and routing packets in a network are examples of such sequential decision tasks. Decision policies for such tasks are difficult to design by hand, but it is often possible to learn them through interaction with the environment. Neuroevolution, where neural networks are evolved with genetic algorithms, is a new and powerful method for learning such policies. In this talk, I will review recent advances in neuroevolution methods, and present several applications ranging from rocket control and autonomous vehicles to robotics and interactive video games.

Risto Miikkulainen is a Professor of Computer Sciences at the University of Texas at Austin. He received an M.S. in Engineering from the Helsinki University of Technology, Finland, and a Ph.D. in Computer Science from UCLA. His current research includes models of natural language processing, self-organization of the visual cortex, and evolving neural networks with genetic algorithms. Professor Miikkulainen is an author of over 180 articles in these research areas, and the books "Computational Maps in the Visual Cortex" (Springer, 2005), "Lateral Interactions in the Cortex: Structure and Function", and "Subsymbolic Natural Language Processing". He is an editor of the Machine Learning Journal and Journal of Cognitive Systems Research.

Date Speaker Title
25/7/2005 Dr. Steve DiPaola, Simon Fraser University Research in intelligent character, creature and artistic interactive systems
Steve DiPaola will discuss and demonstrate his research and the research of the iVizLab (ivizlab.sfu.ca/research) that he directs, which uses an interdisciplinary approach to create human centered interactive, creative and communication systems. Best known for his expertise in 3D facial animation and communication systems, DiPaola will demonstrate iFace which uses an intelligent behavior based parameterized approach to synthetic facial communication. He will also show the lab's ongoing work with the Vancouver Aquarium to create an Artificial Life based (digitalbiology.com), Virtual Beluga Whale Interactive where visitors can collaboratively interact with a simulated pod of wild beluga whales. He will also discuss his work with creative AI systems, that browse face-space for the Maxis game 'The Sims', create moving paintings by extracting emotion out of music and more recently use an automatic creative fitness function to evolve abstract portrait painter programs using Cartesian Genetic Programming (dipaola.org/evolve).
Date Speaker Title
10/5/2005 Prof. Danny van Noort, Seoul National University DNA Computing
Information processing can be seen in various aspects of our life. Actually, nature can be seen an entity where information is processed. DNA is a good example; it is the template of life. This talk will focus on a more basic level where it will be shown how DNA molecules can be used to do computations.

Microflow reactors are an important tool for realizing computing with DNA, RNA or proteins. The advantages are, for example, the need for very small volumes of biological solutions and increased reaction rates. Furthermore, they provide control over the flow of solutions, i.e. the flow of information. Microflow reactors can be made from problem specific to re-configurable systems. A designed DNA strand contains bit information in the sequence of its nucleotides. Extraction of single-stranded DNA molecules, from a sequence space {Si} (the DNA-library), can be made by using complementary capture probes (CP), a short-single DNA strands. Hybridisation between these two is a selection, i.e. a YES or NO.

However, on cellular level it is also possible to do computations. Think of metabolic pathways as a series of logical switches triggered by inhibitors and promoters. This can be seen as a computational network as well. Furthermore, RNA can be used to program cells for desired outputs, like phage. This talk will guide one through various aspects of computation in microreactors anda little inbiology.
Date Speaker Title
28/1/2005 Dr. Jonathan Mackenzie, University of Cambridge Evolving models of morphogenesis in plants
Date Speaker Title
26/8/2004 Dr. Tim Hutton, UCL Making a Self-Reproducing Cell in a Two-Dimensional Artificial Chemistry
Nobody knows how to make an artificial life system that undergoes the evolutionary growth of complexity. This is a shame because if we had a working model we could learn a lot about life and evolution through experiment.

An artificial chemistry is a simplification of real chemistry, designed to be much cheaper to simulate on computer.

In the talk I will show how it is possible to make a self-reproducing cell in an artificial chemistry by surrounding a replicating molecule with a semi-permeable membrane. The molecule can carry an arbitrary amount of information, encoded in a material form as a sequence of bases, as in DNA. The cells produce enzymes through a decoding of their base sequence, and these enzymes trigger reactions essential to the cell's survival.

The cells reliably reproduce over many generations under environmental pressure for resources. By creating cells in a material-based artificial chemistry we hope that the system might have the potential for open-ended, creative evolution.
Date Speaker Title
15/7/2004 Prof. Nik Kasabov, Auckland University of Technology Evolving Connectionist Systems and Applications in Bioinformatics and Brain Study
This presentationgives the background theory of the Evolving Connectionist Systems (ECOS) along with their new development and applications in Bioinformatics and Brain study. ECOS are neural network models that evolve their structure and functionality through learning from data in an on-line and off-line incremental modes, in both supervised and unsupervised modes, and facilitate rule extraction and rule manipulation. The evolving process of an ECOS is defined by parameters, “genes”. ECOS extend further the classical knowledge-based neural networks. Simple ECOS use parameters that have vague analogues with genes in the biological sense. These systems have been applied to challenging problems in Bioinformatics, such as: DNA sequence analysis, microarray gene expression profiling of cancer, finding gene regulatory networks (GRN), medical prognostic systems [1]. Recently new, biologically plausible, ECOS have been developed as computational neurogenetic models, where the ECOS parameters correspond to real genes expressed in the brain and related to learning processes, or to brain diseases. A dynamic model of a GRN within each neuron is evolved during the modeling process that governs the neuronal processes. All neurons have a spiking behavior to form spiking neural networks, characterized by a spectral profile [3]. This model is applied on problems of brain modeling that include modeling learning processes and epilepsy. Further directions for biologically plausible ECOS include: dynamic, evolving neurogenetic models; modeling other brain processes and diseases; general applications of ECOS as universal classification and prognostic systems; self-replicating ECOS; hardware implementation.
Date Speaker Title
6/5/2004 Dr. Robert Smith, University of the West of England Agents, evolution and the end of chaos
The role of “AI” is shifting from dreams of automata that generate optimized results to all manner of problems, to the reality of computers and human beings in creative, collaborative relationships within everyday organizations. This talk begins by considering agent-based models of these relationships. It introduces a view of agents that combines the 19th century views of Peirce with the most recent work of Kauffman. The talk then introduces the novel view that all manner of multi-agent systems can be seen as evolutionary in character; with birth, death, mutation, and the exchange of gene-like and meme-like information between entities. The view will show how the lessons of evolutionary computation can be interpreted and utilized within the context of complex, multi-agent systems. Given this view, the talk will proceed to consider how lessons from complex adaptive systems theory may create control laws for such systems in the future. Final comments consider how such observations may influence the creative collaborations of humans and computers in the future.