AI - The art and science of making computers do interesting things that are not in their nature.
Editing this newsletter is a spare-time occupation, and interesting papers pile up faster than I can read them. This would probably happen even were I not spending time on anything else. Isaac Asimov explained in his essay The Sound of Panting that the panting wasn't him chasing pretty research students around the room, but merely the sound of trying to keep up with the literature. A much more futile task; and that was in the 1950s. So the links below are to papers and topics that I have only skimmed. But I think they're interesting, and perhaps even important.
SlugBot: Towards True Autonomy, by Ian Kelly, 2003:
"Most mobile robots are not truly autonomous; most operate in simplified environments. … On the other hand, even the simplest animals are self-sufficient, both in terms of information processing and energy. The aim of this project is to build a robot with animal-like self-sufficiency in both information and energy. We don't expect to be able to match the speed and performance of a cheetah chasing a zebra, within the time frame of this project, so we decided to chase something slightly slower ... slugs. Apart from their relative ease of capture (compared to zebras), slugs were chosen because they are a major pest, are reasonably plentiful, have no hard shell or skeleton, and are reasonably large. It is also more technologically interesting to catch mobile prey rather than just grazing on plants".
Energy Autonomy: Towards a truly Autonomous Robot.
Description and videos of EcoBot II, whose energy comes from bacterial decomposition of dead flies and rotten fruit in microbial fuel cells.
Robots and autonomy: The EcoBot project.
Chemical details of how the fuel cells derive electricity from bacterial metabolism.
Self-sustaining killer robot creates a stink, by Duncan Graham-Rowe, New Scientist, 9th September 2004:
"In its present form, EcoBot II still has to be manually fed fistfuls of dead bluebottles, but the ultimate aim of the UWE robotics team is to make the droid predatory, using sewage as a bait to catch the flies. … 'One of the great things about flies is that you can get them to come to you,' says Melhuish. The team has yet to tackle this, but speculates that it would involve using a bottleneck-style flytrap with some form of pump to suck the flies into the digestion chambers. … In tests, EcoBot II travelled for five days on just eight fat flies - one in each MFC [microbial fuel cell]".
REACTION-DIFFUSION COMPUTERS: 5th International Conference on Unconventional Computation 4-8 September 2006, University of York, UK.
This ad for a tutorial on reaction-diffusion computing explains this to be: "computation with travelling waves in reaction-diffusion non-linear media. A reaction-diffusion computer is a massively parallel computing device, where micro-volumes of the chemical medium act as elementary few-bit processors; and chemical species diffuse and react in parallel. In the reaction-diffusion computer both the data and the results of the computation are encoded as concentration profiles of the reagents, or local disturbances of concentrations, whilst the computation per se is performed via the spreading and interaction of waves caused by the local disturbances".
Reaction-diffusion computers: Wet, Weird and Wonderful, by Andrew Adamatzky and Ben De Lacy Costello.
A two-page summary of reaction-diffusion research. Potential applications include image processing, planning, robot navigation, and computational geometry; and implementation of novel logic gates and storage devices.
Programming Reaction-Diffusion Processors, by Andrew Adamatzky, 2005.
Research paper on three ways of programming reaction-diffusion processors: one computes Voronoi diagrams, two implement logic gates. Includes references to previous work.
Adamatzky's home page. Links to a great number of publications. Frustratingly, most are not on the Web.
Greg Turk's page on creating surface texture in computer graphics by simulating reaction-diffusion.
RESEARCH IN IMMUNOCOMPUTING: Dr. Alexander Tarakanov.
Near the bottom of this page is a fuzzy but tantalising screen shot showing clusters of blobs, one labelled "Molniya-1 N67" and the other "Shuttle Ferret". To quote the text: "We develop a rigorous mathematical basis of a novel approach to computing, immunocomputing, and its applications to solve specific real world problems. Immunocomputing is inspired by the biological principles of information processing by proteins and immune networks. We introduce new mathematical abstractions of formal protein and formal immune network. We developed a rigorous proof that a formal immune network is able to learn, to recognize, to solve problems and to represent languages based on the theory of linguistic valence. We present some applied results, such as computing of ecological atlases, monitoring of most dangerous infections, detecting critical situations in near Earth space, information security, etc. These results allow us to speak about the hardware implementation of the immunocomputing in so-called immunochip. We have developed several versions of software emulator of the immunochip. Currently we are developing a biochip (biological microchip) for immunocomputing".
INFORMATION SECURITY WITH FORMAL IMMUNE NETWORKS, by Alexander Tarakanov.
I wrote about some applications of artificial immune systems in our March issue. Despite curiosity about the images above, I didn't mention Tarakanov's work, partly because almost all his papers are on pay-to-view sites. This one isn't. It also looks more rigorous than some other AIS research, and therefore worth following up.
Immunocomputing for Surveillance the Plague in Central Asia, by Alexander Tarakanov and Svetlana Sokolova.
Another of Tarakanov's few freely-available papers, modelling the dynamics of epidemic disease in central Asia. The authors conclude that immunocomputing can focus attention on the most dangerous situations, something beyond the possibilities of traditional statistics.
Nonmonotonic Inferences and Neural Networks, by Reinhard Blutner. Marked as to appear in Knowledge, Rationality and Action, Issue 2, 2004.
It's good to see work that tries to unify separate areas of research. Blutner demonstrates an equivalence between non-monotonic inference, and Hopfield nets considered as dynamical systems: "The main results are (a) that certain activities of connectionist networks can be interpreted as nonmonotonic inferences, and (b) that there is a strict correspondence between the coding of knowledge in Hopfield networks and the knowledge representation in weight-annotated Poole systems. These results show the usefulness of nonmonotonic logic as a descriptive and analytic tool for analyzing emerging properties of connectionist networks. Assuming an exponential development of the weight function, the present account relates to optimality theory - a general framework that aims to integrate insights from symbolism and connectionism. The paper concludes with some speculations about extending the present ideas".
Category Theory Applied to Neural Modeling and Graphical Representations, by Michael Healy. Minor revision of Paper NN0648 in International Joint Conference on Neural Networks 2000, Como, Italy.
In programming, functions can be defined in terms of other functions; lists can have sublists as elements; records can have subrecords as fields; classes can inherit from subclasses. The resulting functions, lists, records, or classes can then be further combined and extended … and so on, as many times as required. Such recursive extensibility is rare in neural nets, but appears to be what Healy is trying to provide. He uses category theory's "colimit" operation, a kind of generalised sum. This builds a description of a system from descriptions of its components — program modules, database views, processes, or whatever — given information about how these are interconnected. Importantly, it works when different components share information, as in two overkapping database views, or two program modules both of which share a submodule. Or, as Healy has, two neural nets that gain different views of a concept through different sensors.
The categorical model, with functors from a category of concepts to a category of neural network components and natural transformations between these functors, provides a mathematical model for neural structures consistent with concept-subconcept relationships. Colimits of diagrams show how concepts can be combined, and how a concept can be re-used many times in forming more complex concepts. Functors map commutative diagrams to commutative diagrams, capturing this aspect of the colimit structure. Natural transformations express the fusion of single-mode sensor representations of concepts in the same erent implementations of the concept hierarchy at all levels and in a consistent fashion. This mathematical model appears to be compatible with a model of the primate brain proposed by Damasio. We hope that others will find this model interesting and useful for neural network analysis and design.
Brief slide presentation on Category Theory and Cognitive Neural Systems: A Mathematical Semantic Model, by Michael Healy and Thomas Caudell, 2004. Note the "association area" between two nets in the diagram.
Style as a Choice of Blending Principles, by Joseph Goguen and D. Fox Harrell.
This paper applies category theory too, to formalise how the meaning of a metaphor is derived from its components. The components are the meanings of words such as "house" and "boat"; the result is an optimal blend of these, as in the word "houseboat". In a Cognitive Linguistics List announcement, Goguen says this work offers:
Formal Notation for Conceptual Blending, by Joseph Goguen.
"This note concerns formal notation for conceptual spaces, conceptual morphisms, and conceptual blending, in the sense developed in An Introduction to Algebraic Semiotics, with Applications to User Interface Design, … Here, we expand and correct certain aspects of the discussion there; also, instead of the notation developed there, we use the notation of the OBJ algebraic specification language, which is executable, though we use that capability mainly to check the syntax of our theories and morphisms, i.e., as a type checker for algebraic semiotics. For much more detail on OBJ, see the manual, Introducing OBJ.
Derek Tastes of Earwax, from the BBC Science & Nature: TV & Radio Follow-up for Horizon.
"Imagine if every time you saw someone called Derek you got a strong taste of earwax in your mouth. It happens to James Wannerton, who runs a pub. Derek is one of his regulars. Another regular's name gives him the taste of wet nappies. For some puzzling reason, James's sense of sound and taste are intermingled". The page, which links to tests anyone can do, and some questions and answers about synaesthesia, continues with evidence suggesting that we are all synaesthetic to some extent.
Lecture 4: Purple Numbers and Sharp Cheese, from the 2003 Reith Lectures by Vilayanur Ramachandran.
Ramachandran desscribes synaesthesia in some detail, and explains the kiki-booba test. This demonstrates one form of "synaesthesia" — of cross-modal neural association — that almost all of us have.
The Science Behind the Story: The Skeekit-Woogle Test, by Carl Frederick, concerning his story about synaesthesia in Analog. Same test as Ramachandran's; different name.
Hearing colours at the symphonie fantastique, by Julian Asher, in The Colourist, Summer 2003.
An account of coloured-music synaesthesia and of the first large-scale study on it, made possible by CDs.
Synaesthesia - A Cognitive Model of Cross Modal Associationm by Andrew Lyons, Composition Unit, Sydney Conservatorium of Music, University of Sydney.
A short paper with drawings of the "photisms" seen by coloured-hearing synaesthetes. These apparently resemble "Kluver's form constants", relatively simple shapes also seen in hallucinations and "primitive" art.
Everyday fantasia: The world of synesthesia, by Siri Carpenter, Monitor on Psychology, Volume 32, Number 3, March 2001.
Popular-science account of some imaging studies and proposed neural mechanisms. The latter are not yet detailed: psychologist Peter Grossenbacher from Naropa University is quoted as saying "The trouble with theorizing in this area is that we're underconstrained by data. There isn't the right kind of data, yet, to differentiate between these different theories".
[ Jocelyn Ireson-Paine's Home Page ]