Tuesday, May 31, 2005
GA-fuzzy researcher named dean at Alabama
Soccer simulation with GAs and agents
Saturday, May 28, 2005
Sailing season and genetic algorithms
A quick Google search on yacht and genetic algorithms turns up many interesting links, but the first article on this topic I remember was some pretty cool work by Carlo Poloni. The study used multiobjective GAs and neural-net surrogate evaluation to optimize yacht design. A pdf of a 1999 presentation is here. I believe that Carlo's work on multiobjective GAs has formed the basis for design optimization software at an Italian firm called Enginsoft.
Thursday, May 26, 2005
I've been using Electric Sheep in my screensaver rotation for over a year, and I just now realized that it is a form of collaborative GA-based art.
I strongly recommend folks check it out. Warning, it will hypnotize the unsuspecting.
Let The AI Blog For You!
Make sure and check out the "real stories" of how it's generating most of the blogs everyone reads these days.
(Could Goldberg be using Autoblogger? Surely he can't have that much free time on his hands!)
AI = BS
There is something really beguiling about human Vs computer chess matches.
Despite the main elements being a bloke who plays chess and a very complicated bit of software, these kind of matchups somehow transcend simple geekiness.
Maybe we feel there's something lurking in these tournaments that holds the key to the future of the human consciousness: a profound kernel of truth that will maybe give us some clue to what lies at the end of that mysterious road marked 'Evolution'.
Well, and maybe not.
I'm often described as an "AI guy". But I actually hate the term artificial intelligence. It's largely indicative of the misplaced idea that rationale is the most unique and prized quality of man. Rubbish.
Chess program AI misses the point of the pursuit which most so-called AI guys are following. Dolphins, bee hives, and your slow-witted Uncle Jeeter don't play chess, yet they most certainly have the ephemeral quality that we deeply desire for our artificial creations.
I think that as our machines begin to show a familiar, yet perpetually novel complexity of interaction in our social context, they may begin to have the quality that most AI guys are really looking for.
I have no good ideas for alternate acronyms, though. Artificial friends and neighbors, perhaps?
Wednesday, May 25, 2005
Live from Shanghai II: ECN Workshop
Stochastic local search
Evolutionary self-replicating machines
The Cornell machines, dubbed Molecubes, measure 10 centimeters to a side and are split diagonally. Each cube half swivels on a motorized axel in 120-degree increments. The cube faces have electromagnets that strengthen and weaken to make and break connections with other cubes, and contacts that transfer communications and power between cubes.
The machines are powered through a base plate mounted on the floor of their enclosure, and they receive new cubes that the researchers place by hand in specific locations. Stacks of three and four cubes can assume a variety of shapes and, by following rules governing when and how to move after each contact with another cube, three- and four-cube machines can build copies of themselves. A three-cube machine takes just over a minute to reproduce; a four-cube machine takes two and a half minutes.
The researchers have also produced software simulations that show that self-replication is possible with larger numbers of cubes. The simulations were of seven- and eight-cube machines whose shapes and controllers were generated by an evolutionary algorithm
Now all that is to be done is build functional humanoid robots that reconfigure and repair themselves and say "asta la vista baby!"
Tuesday, May 24, 2005
Live from Aguascalientes
AAAI Fall Symposium Hosts Symposia on Evolutionary, Adaptive and Anticipatory Systems and Mechanisms
Hierarchical BOA for military antenna design
Recently, an application of hBOA to the military antenna design caught my interest where the task is to optimize a novel, wideband overlapped subarray system to achieve -30-dB sidelobes over a 20% bandwidth. This work was done by the Illinois Genetic Algorithms Laboratory (IlliGAL) and the Air Force Research Laboratory (AFRL). A number of researchers from IlliGAL and AFRL were involved in this collaborative project: Scott Santarelli, Tian-Li Yu, David E. Goldberg, Edward Altshuler, Teresa O'Donnell, Hugh Southall, and Robert Mailloux. The results show that while simple genetic algorithms fail to solve this problem, hBOA provides acceptable solutions as the problem difficulty increases. The results thus confirm that while for simple problems simple genetic algorithms may suffice, difficult problems necessitate the use of more advanced, competent, genetic algorithms.
For more information, check out IlliGAL report 2005013.
Genetic algorithms for distributed data storage systems
Looking for ways to ensure data stays intact on an experimental network, the group found that genetic algorithms provided a handy shortcut in determining how to distribute data across multiple machines, according to Leana Golubchik, a professor of computer science at the University of Southern California. Golubchik, who is a co-researcher on the project, presented the technology at this year's NSF-sponsored Digital Government Research conference held this week in Atlanta.
The findings stem from a data transfer technology called Bistro that the researchers have been working on for the past few years with support from NSF. Bistro would allow agencies to accept large amounts of data arriving simultaneously from many sources. Such a service could prove useful to agencies such as IRS, which sees an influx of material during tax deadlines.
Monday, May 23, 2005
Advice to my penguins: Learn Chinese
Live from Shanghai
Call For Participation in the PSGEA-2005 Workshop at GECCO-2005
The workshop will start with an introduction by the organizers, followed by 6 talks, and will finish with a panel discussion. Attendance to the workshop is open to all GECCO attendees.
List of talks given at the workshop:
- Population Sizing for Genetic Programming Based Upon Decision Making
K. Sastry, U.-M. O'Reilly, and D. E. Goldberg
- Parameter Sweeps For Exploring GP Parameters
M. E. Samples, J. M. Daida, M. Byom, and M. Pizzimenti
- Genetic Programming: Parametric Analysis of Structure Altering Mutation Techniques
A. Piszcz and T. Soule
- Investigations in Meta-GAs: Panaceas or Pipe Dreams?
J. Clune, S. Goings, B. Punch, and E. Goodman
- A Review of Adaptive Population Sizing Schemes in Genetic Algorithms
F. G. Lobo and C. F. Lima
- Online Population Size Adjusting Using Noise and Substructural Measurements
T.-L. Yu, K. Sastry, and D. E. Goldberg
Sunday, May 22, 2005
On way to Shanghai
Friday, May 20, 2005
DISCUS article picked up by HPCwire
Thursday, May 19, 2005
Wednesday, May 18, 2005
DISCUS makes headlines at NCSA news
Tuesday, May 17, 2005
Cool conference innovation
but we did have the very interesting --in fact compelling—use of the projection of the forum chat room on the wall behind us. This made the audience a real participant in the conversation, especially the snarkiest of commentators. This is an evolution in conferences, and I urge its immediate adoption. If you don’t like being mocked by a live audience in real time, then don’t be a talking head. This innovation sure made for an interesting backdrop to the standard five microphones and a few hundred people in the audience panel.Ooh, I like the idea of conference chats (and blogs?) going on during a conference, and then projecting them at the back of a panel discussion is even better. Might be a great way to spice up a GECCO workshop.
Saturday, May 14, 2005
What is General Engineering (GE)?
During the Cold War, the GE degree took on a systems engineering flavor, and an MS degree was added in the 70s. More recently, a PhD has been added, but a PhD in General Engineering sounded like something of an oxymoron, so the PhD offered was called Systems and Entrepreneurial Engineering (SEE). This degree is something of a cross of offerings in Management Science and Engineering (MS&E) at Stanford and the programs in the Engineering Systems Division (ESD) at MIT.
The undergraduate program in General Engineering is one of the most popular at Illinois with nearly 600 undergraduate students. The new SEE degrees have attracted applicants from around the world, and is growing rapidly. More information is available on the GE website here.
Top 5 GA programs. . . in Brazil
Educating a penguin, part III
Blogging from GPTP
GAs used to study protozoan glyoxalase pathway
The glyoxalase pathway of Leishmania infantum was kinetically characterized as a trypanothione-dependent system. Using time course analysis based on parameter fitting with a genetic algorithm, kinetic parameters were estimated for both enzymes, with trypanothione derived substrates. A Km of 0.253 mM and a V of 0.21 µmol·min–1·mg–1for glyoxalase I, and a Km of 0.098 mM and a V of 0.18 µmol·min–1·mg–1 for glyoxalase II, were obtained.
The research was a collaboration between Portugese researchers at the University of Porto and the University of Lisbon.
GP in SQL
Thursday, May 12, 2005
IEC at WSTST2005
Interesting feature is that four of them discuss how to reduce IEC user's fatigue, while the number of IEC application papers is much bigger than that of IEC interface papers in past conferences.
Evolving Computer Viruses
Wednesday, May 11, 2005
25 pieces of advice for bloggers
If you're going to be putting up multiple posts and then not posting for a while, put the best post on top. The number of readers drops significantly the further they have to go down the page.So does 4:
On the week-ends, expect your traffic to drop by roughly 40% whether you post or not.
A previous post about Hugh Hewitt's book Blog has a better list for getting started I think.
Monday, May 09, 2005
New Illigal Technical Reports
Butz, M.V., Pelikan, M., Llorà, X., Goldberg, D.E. (2005). Extracted Global Structure Makes Local Building Block Processing Effective in XCS. IlliGAL Report No. 2005010. (Abstract) (Full paper in PS) (Full paper in PDF)
Butz, M.V., Pelikan, M., Llorà, X., Goldberg, D.E. (2005). Automated Global Structure Extraction For Effective Local Building Block Processing in XCS. IlliGAL Report No. 2005011. (Abstract) (Full paper in PS) (Full paper in PDF)
Lanzi, P.-L., Loaicono, D., Wilson, S. W., Goldberg, D. E. (2005). Generalization in the XCSF Classifier System: Analysis, Improvement, and Extension. IlliGAL Report No. 2005012. (Abstract) (Full paper in PS) (Full paper in PDF)
Santarelli S., Yu, T.-L., Goldberg D. E., Altshuler E., O’Donnell T., Southall H., Mailloux R. (2005). Military Antenna Design Using Simple and Competent Genetic Algorithms. IlliGAL Report No. 2005013. (Abstract) (Full paper in PS) (Full paper in PDF)
Yu, T.-L., Yassine, A., Goldberg, D.E. (2005). An Information Theoretic Method for Developing Modular Architectures Using Genetic Algorithms. IlliGAL Report No. 2005014. (Abstract) (Full paper in PS) (Full paper in PDF)
Yassine, A., Goldberg, D.E., Yu, T.-L. (2005). Simple Models of Hierarchical Organizations. IlliGAL Report No. 2005015. (Abstract) (Full paper in PS) (Full paper in PDF)
Yu, T.-L., Sastry, K., Goldberg, D.E. (2005). Linkage Learning, Overlapping Building Blocks, and a Systematic Strategy for Scalable Recombination. IlliGAL Report No. 2005016. (Abstract) (Full paper in PS) (Full paper in PDF)
Yu, T.-L., Sastry, K., Goldberg, D.E. (2005). Online Population Size Adjusting Using Noise and Substructural Measurements. IlliGAL Report No. 2005017. (Abstract) (Full paper in PS) (Full paper in PDF)
Other IlliGAL technical reports and publications are available here.
Friday, May 06, 2005
Blogging memetic algorithms
Having said that, I am fond of hybrids for many if not most industrial-strength GA applications. We've worked a bit on the theory of local-global hybrids (see post here and tech report here), and more needs to be done, but in practice hybrids are a good way to get the broad perspective of a GA together with the local convergence speed of a domain-appropriate local searcher.
Symbiot uses GAs for adaptive network security
Symbiot utilizes proprietary genetic algorithms to measure, manage and mitigate risk to your networked assets. Through Symbiot.NET, Symbiot's customers benefit from adaptive profiles defined from industry groups, and other Symbiot customers for community centric security; a new approach to mitigating risk. Symbiot provides solutions that unify your existing security infrastructure to proactively respond to business critical issues while communicating security events across your entire organization clearly and effectively.
A number of technical white papers are available here.
Registration up: Biggest, best GECCO ever?
Thursday, May 05, 2005
NuTech Solutions raises funds
Moneyscience picks up on OBUPM Workshop
Top five GA programs?
El mundo de adan posted a list of schools with good programs in genetic algorithms. Here is the list with El mundo de adan's comments:
- university of rochester (not so much ga, but good ai in general, plus it's
obviously in rochester)
- syracuse univeristy
- michigan state
- colorado state
- university of edinburgh (scotland!)
Someone wanting to stay in Rochester might talk to Al Biles at Rochester Institute of Technology. Someone wanting to stay in the States, hmmmmm (let me think about that), maybe, just maybe might consider (drum roll please) Illinois! Although IlliGAL is in a department called General Engineering, most of the PhD students are in CS, and the little ole director has an affiliates appointment in CS.
We leave it as an exercise to IlliGAL Blogging readers to add their favorite nominations for GA-friendly schools in the US and around the world. Go ahead and give your top five in the comments section.
Oops! Email lowers IQ more than marijuana
This raises the unsettling possiblity that perhaps the postmodern equivalent of walking and chewing bubble gum has become IMing and smoking a joint. Hat tip to IFTF Future Now.
In 80 clinical trials, Dr. Glenn Wilson, a psychiatrist at King's College London University, monitored the IQ of workers throughout the day.
He found the IQ of those who tried to juggle messages and work fell by 10 points -- the equivalent to missing a whole night's sleep and more than double the 4-point fall seen after smoking marijuana.
"This is a very real and widespread phenomenon," Wilson said. "We have found that this obsession with looking at messages, if unchecked, will damage a worker's performance by reducing their mental sharpness.
Wednesday, May 04, 2005
Towards Distinguishing Levels of Consciousness
In reply to Goldberg’s post on “Towards Conscious Machinery”, I will discuss herein what we might want to term conscious and, more importantly, which states of consciousness might be worth distinguishing.
Goldberg agrees to the general conjecture that we need a body and an interactive system that is able to manipulate and perceive the own body, as well as perceive manipulations by other entities in the environment. Goldberg does not agree to the need for symbolic-like representations or other further structuring capabilities for consciousness experience.
I believe that we may need to distinguish the type of consciousness we are referring to, in order to be able to further discuss, which mechanisms are mandatory for conscious experience.
For the lowest level of machine consciousness, the one we might be willing to attribute to our dog-and possibly most other mammals and potentially even birds (see e.g. a short article in Nature, 430, 414-414 (22 Jul 2004) on the social capabilities of birds), language capabilities are certainly not necessary. But it needs to be asked, which levels of consciousness are these animals able to reach and what is the fundamental difference between such probable lower-level consciousness experiences and our more expressible experiences of consciousness?
O’Regan and Noë provide a potential answer in their article A sensorimotor account of vision and visual consciousness published in Behavioral and Brain Sciences, 2001, 24(5), 939-1011. For them, feelings arise out of the sensory-motor contingencies that everybody experiences in everyday life. That is, the continuous anticipation of what may come next and the continuous experience of what did actually come next (that is, how the world changed – partially induced by our own actions) results in the experience of feelings and low-level consciousness for that matter. Clearly, such feelings then are reflected upon in dreams (including dreams in mammals and potentially birds) so that this feeling may be sufficient to be termed consciousness in the Searlean sense of "those states of sentience or awareness that typically begins when we wake up from a dreamless sleep", as Goldberg cited.
However, higher states of consciousness require more abstract processing capabilities in my opinion. As also pointed out in the above mentioned article on bird intelligence, it is probably not sufficient to be conscious and reflective without a significant social component such as the capability of distinguishing other individuals – effectively improving the interaction and overcoming an (iterated) prisoner’s dilemma by remembering and distinguishing the different interactions with others. To do so, more abstract symbolic like processing is mandatory tagging other individuals with behavioral properties. The distinction then may lead to the more concrete perception and internal representation of oneself being yet another (certainly with the very special property of selfness) individual.
Language capabilities are yet another stage in this process that enable us to gain even more abstract representations and allows us to reflect upon ourselves and our environment on much higher levels of abstraction in time (predictive) and space (object and modular oriented). Hereby, it is important that language is yet another level of abstraction – grammar emerges out of the coevolution of social language, our environment, and our brain capabilities as Deacon points out.
Tuesday, May 03, 2005
Invitation to the OBUPM-2005 Workshop at GECCO-2005
OBUPM-2005 is a half-day workshop organized by Joern Grahl (University of Mannheim in Germany), Kumara Sastry (University of Illinois at Urbana-Champaign), and Martin Pelikan (University of Missouri at St. Louis). Participation at the workshop is open to all GECCO-2005 attendees. The workshop will contain 7 presentations:
- An Overview of the Extended Compact Genetic Algorithm
Authors: Georges Harik, Fernando Lobo, Kumara Sastry
Presenter: Fernando Lobo
- Approximate Factorizations of Distributions and the Minimum Relative Entropy Principle
Authors: Heinz Mühlenbein and Robin Höns
Presenter: Heinz Mühlenbein
- A Survey to Probabilistic Model Building Genetic Programming
Authors: Yin Shan, Robert McKay, Daryl Essam, Hussein Abbass
Presenter: Hussein Abbass
- Efficiency Enhancement Techniques in Estimation of Distribution Algorithms
Authors: Martin Pelikan, Kumara Sastry, and David E. Goldberg
Presenter: Martin Pelikan
- Scalability of Multiobjective Estimation of Distribution Algorithms
Authors: Kumara Sastry, Martin Pelikan, and David E. Goldberg
Presenter: Kumara Sastry
- Numerical Optimization with Real-Valued Estimation-of-Distribution
Authors: Peter A. N. Bosman
Presenter: Peter A. N. Bosman
- Optimization of a Constrained Feed Network for an Antenna Array Using Simple and Competent Genetic Algorithm Techniques
Authors: Tian-Li Yu, Scott Santarelli and David E. Goldberg
Presenter: Tian-Li Yu
Most of the presented material (and a lot more than that) will be published with Springer in an edited book on estimation of distribution algorithms, which will cover most important advances in this field (book editors: Erick Cantu-Paz, Kumara Sastry, and Martin Pelikan).
Monday, May 02, 2005
Educating a penguin: Part II
This post is not a tell-all confession by a large-university insider, but it is no secret that large research universities emphasize research. That small liberal arts schools might better educate undergraduates should come as no surprise. That statistical studies of PhD productivity and Who's Who prominence tilt in favor of a number of largely unheralded liberal arts schools were news to me.
As a result, Max and I need to get back on the road, and included in our visits will be some of these apparently life-transforming institutions. Where Max goes will be his choice, but we need to take a more informed look off the beaten track.
Sunday, May 01, 2005
Some of the new AI research also falls into an emerging niche of computer science: the intersection of artificial intelligence and human-computer interaction.Exactly. Trying to balance human-computer interaction is a key goal of DISCUS. The 4-quad chart (here) captures some of the possiblities in the space of choosing and creating.
Creationism and genetic algorithms
I've been frankly startled by the power and efficacy of the evolutionary process for solving problems, without even stating the nature of the problem itself to the solution engine. Watching powerful solutions arise out of the 'digital goo' in real time is downright spooky. And that's with relatively tiny population sizes.
But I'm not sure that anecdotal observation of GA performance is all that helpful in "settling" anything. If fossil records and increasingly detailed undestanding of genetics and cellular function down to the molecular level aren't persuasive, should we expect the caricature simulations of GAs and GP over evolutionary time scales to really turn the tide?
Having said this, understanding GA/GP time complexity, problem class appropriateness, and solution quality (see DoI) may be able to put to rest unsophisticated arguments equating natural evolution to simple random search. Moreover, these complexity arguments (with a good bit of work) might yield a detailed bound on the plausibility of the complexity of natural systems we now observe.
Tissue classification using GAs
Several machine learning approaches have been used to aid to understand the functions of genes. However, these tasks are made more difficult due to the noisy nature of array data and the overwhelming number of gene features. In this paper, we use the parallel genetic algorithm to filter out the informative genes relative to classification. By combing with the classification method proposed by Golub et al. and Slonim et al., we classify the data sets with tissues of different classes, and the preliminary results are presented in this paper.
The full text of the paper is available here.