Dear my friends in the world,
I made this 2005 annual report as my seasonal greeting to my friends. Please enjoy it.
I hope that we keep in touch
Thinking of you from Japan on New Year Day, 2005
--- My topics in 2004 ---
During the 2004 fiscal year (April, 2004 - March, 2005), Takagi Laboratory consists of two Bachelors, three Master students, and one doctoral student. But, a part-time Master student and a doctoral student are taking off university, so that only three students work for our lab actually.
In the 2005 fiscal year, four Master students and a Governmental scholarship student who will challenge the enrollment to doctoral course from Slovakia will enroll to our lab in April. The number of Bachelors to our lab from April is unknown so far, and how students taking off university will do from the 2005 fiscal year, too, but Takagi Lab will consists of seven students at least.
The biggest difference of Takagi Lab this year from past years is the number of postdoctoral researchers. Three postdoctoral researchers will support our research activity strongly: Dr. Shangfei Wang, COE postdoctoral researcher in July, 2004 - December, 2005, Dr. Raffi R. Kamalian, JSPS Postdoctoral Fellow in November, 2004 - November, 2006, and Dr. Shino Iwashita, March, 2005 - March, 2006. Postdoctoral researchers hired by Kyushu University have one-year contract, but we can expect their strong support in the 2006 fiscal year if they renew the contracts of employment.
I hosted Lecture Huidao Tang from China and Prof. Oswaldo Velez-Langs from Spain as Visiting Researchers in 2004.
I have been working on the combination of human factor and computational intelligence that I call as Humanized Computational Intelligence. Interactive evolutionary computation (IEC) is still a main keyword of my research for this direction. See a survey and tutorial paper of IEC research.
We, team members who are applying to governmental Grants-in-Aid for Scientific Research (not accepted yet, though), had symposiums on Humanized Computational Intelligence at Spa in January and August, 2004, and will have the third symposium in January, 2005. We had and will have longer discussion time more than presentation time in relax emotion in these symposiums and stayed/will stay a night together at a Spa hotel in Japanese style. We hope that our grant application will be accepted in 2005. I also hope that I can find research direction on Humanized Computational Intelligence that exceeds IEC on which I have researched.
Our IEC research in 2004 includes the following.
Extended IEC Based on Physiological Feedback: This is a research directly related with the 21st Century COE Program on "Design of Artificial Environments on the Basis of Human Sensibility. To realize the projective objectives, it is necessary to optimize the parameters of artificial environments, such as heat, lights, vibration, sounds, and visual stimulus, in order to control human physiological conditions for certain human activities such as sleeping, relaxing, or office working. From the system theory point of view, this is an inverse problem that decides inputs for given outputs. We proposed an extended IEC that optimizes a target system based on physiological feedback for the COE project in 2004, while normal IEC is based on the feedback of subjective evaluation. We started to apply the extended IEC to optimize a vibration environment in 2004, but we are going to apply it to audio or visual environmental tasks which are easier experimental tasks for evaluating evaluate the proposed method in 2005. We proposed the extended IEC at COE International Symposium on Design of Artificial Environments held in December, 2004, and will report it as a Rapid Communication in Journal of Physiological Anthropology and Applied Human Science.
VR Presence vs. VR Sickness: This is a doctoral student's work related with the 21st Century COE Project. There is a trade-off relationship between virtual reality (VR) presence and VR sickness; it becomes easier to feel both VR presence and VR sickness according to increase the visual angle and angular velocity of the display to VR users. As the characteristics of this relation depend on users, it is requested to develop a tuning method that maximizes the VR presence within the range of no VR sickness by examining the VR presence-sickness characteristics of each VR user.We introduced two neural networks (NN's), -- an NN that inputs a visual angle and angular velocity of a VR display and outputs the VR presence level (vection occurrence rate) and an NN that inputs same visual angle and angular velocity and outputs the VR sickness level (simulator sickness questionnaire total score), -- and proposed a method to find the best solution by observing the outputs of two NN's trained by a few number of VR examination. We presented this research at COE International Symposium on Design of Artificial Environments held in December, 2004, and will report it as a Rapid Communication in Journal of Physiological Anthropology and Applied Human Science in February, 2005.
Reduction of IEC User's Fatigue: IEC is a system optimization method where an IEC user evaluates the output of the target system and evolutionary computation (EC) optimizes the target system. As this process iterates many times, we cannot avoid a human fatigue problem. One of proposed solutions is to introduce a learning function and to make the function learn the evaluation characteristics of an IEC user. Once a pseudo-human is learned in a computer, we can run the pseudo-human and EC with many numbers of EC individuals behind a real IEC user's evaluation and show the best n individuals to the real IEC user in the order that is easier for an IEC user to compare individuals and evaluate them. Problem of this method is that we cannot use all evaluation data of the IEC user in past. As the IEC user evaluates individuals relatively in each generation, it may happen that an individual with the highest score in the 2nd generation may be poorer than an individual with the lowest score in the 10th generation, for example.Our proposal is to introduce the mechanism that maps the relatively evaluated subjective fitness values to an absolute evaluation scale. We evaluated its performance and presented at COE International Symposium on Design of Artificial Environments held in December, 2004, and will report it as a Rapid Communication in Journal of Physiological Anthropology and Applied Human Science.
IEC for Mental Health Measurement: This is a research to use IEC to measure mental health with a clinical therapist, Mr. Takahashi, for these several years. IEC user’s psychological evaluation scale is reflected in his or her subjective evaluation characteristics, and these in turn, are reflected in the systems designed with IEC. We may therefore observe the psychological scale or the mental health state of the user indirectly by observing the system outputs. This is a quite different approach of normal IEC, which is a system optimization. We had three schizophrenics and five mentally healthy students design happy and sad impression computer graphics (CG) lighting images using IEC and asked other 33 students to evaluate the CG images using Scheffe’s method of paired comparison. Statistical tests of the evaluations showed that the range of emotional impressions perceived by the three schizophrenics between happy?sad was significantly narrower than that which was perceived by the mentally healthy students. We presented this new applicability of IEC in June and October, 2004. We now had additional data and are analyzing the data. We are planning to extend experimental subjects to persons of senile dementia and trouble behaviors and hope that this technique becomes a useful tool in psychiatry.
International Joint Research:
(a) IEC for MEMS Design: Prof. Agogino of University of California at Berkeley and her students
have applied genetic algorithm (GA) to design MEMS (micro-electro-mechanical
system) as a multi-objective optimization task. Their approach is useful
for measurable fitness values but cannot involve domain experts' visual
inspection which is based on their experience and domain knowledge. To
solve this problem, we started to introduce IEC to this task. Her student
came to our lab in summer, 2003, and leaned IEC for 50 days, and we presented
our first result at GECCO2005 in June, 2004. He completed his PhD research
on the GA and IEC-based MEMS design and took his PhD. He is Dr. Raffi R.
Kamalian and became our lab member since last November as a JSPS postdoctoral
fellow for two years.
(b) IEC for Underground Water Management: Prof. Minsker of University of Illinois at Urbana-Champaign and her students have handled underground water management as a multi-objective optimization task using GA, too. They started to introduce IEC for the same reason with the above (a) and stayed at our lab in May and June, 2003. This research was presented in 2004, and we are expanding this research by introducing a fuzzy system into this IEC framework.
Our research publication in 2004 includes one book chapter, three journal papers, five international conference papers, one tutorial paper for a journal, and three domestic conference papers.
My targets in 2005 are:
to experimentally show that extended IEC can optimize a target system with physiological feedback,
to start new research on designing a deep-emotion enhancement filter based on measured physiological data of a human who is impressed by movies or music as an application of the above 1.,
to spread the IEC technique for designing circuit, MEMS, or device including filing patents with UC Berkeley,
to summarize the experimental data of mental health measurement and submit a journal paper, and extend our experimental subjects to persons of senile dementia,
to evaluate the effectiveness of IEC-based medical image enhancement filter design with the help of medical doctors and students, quantitatively,
to design policy and manual for IEEE SMC technical activity and let SMC continue their stable activity in technical aspects, and
to make my own vision and mission for Humanized Computational Intelligence.
I taught courses on computational intelligence, software engineering, data and signal compression, and other lab works in 2004.
I had a chance to give a lecture to undergraduate students who taking a course of System Optimization at BogaziciUniversity in Istanbul on December 16, 2004.
Prof. Iba and I were guest editors for the Special Issues on Interactive Evolutionary Computation at an international journal, New Generation Computing in 2003 and 2004.
The issues will be published in Vol.23, No.2, February, 2005.
I was elected as a Board of Governors member of IEEE SMC Society for 2004-2006.
I also start to serve for whole SMC Technical Committee as the Technical
Committee Coordinator since the end of last year. I resigned the Soft Computing
Chair under the SMC Technical Committee at the end of 2004 after six-year
serve and passed its baton to Prof. Murata of Kansai University who significantly
contributed for organizing Soft Computing Tracks at SMC2002, SMC2003, and
SMC2004. I keep serving as an Associate Editor of IEEE Transactions on
SMC, Part B.
This year is the final year for serving as a SOFT (Japan Society for Fuzzy Theory and Intelligent Informatics) Councilor.
We received a Paper Award from SOFT for the below paper and received it
at SOFT General Meeting at FSS2004 in June, 2004:
Hideyuki Takagi, Takeo Ingu, and Kei Ohnishi,
"Accelerating a GA Convergence by Fitting a Single-Peak Function,"
J. of Japan Society for Fuzzy Theory and Intelligent Informatics, vol.15, no.2, pp.219-229 (2003）(in Japanese).
[ABSTRACT] The simplest approximation of GA landscape is a single-peak function such as a quadratic function or Gaussian function. As the peak position is expected to locate near the global optimum, this method approximates the GA landscape and adds the position of the peak as an elite individual into individuals in the next generation by replacing the poorest individual. If the elite is a better individual as expected, it contributes to accelerate GA convergence; if not, it is just one of many individuals and it may not make the GA convergence worse. We can say that this is a method of "low risk, high return".
Other academic activity in 2004 was giving a tutorial talk at GECCO2004. In 2005, it is scheduled to give plenary talks at WSTST2005 (Muroran, Japan in May), SMCia2005 (Helsinki in June), and KES2005 (Melbourne in September) and a tutorial talk at GECCO2005 (Washington DC in June).
My elder daughter enrolled Kobe Women's College in April, 2004, and younger daughter will take the enrollment examination to High School next month. The younger one will be an exchanging high school student in US for 10 months from this summer. My wife and I will back to the honeymoon days for 10 months.