Dear my friends in the world,
I made this 2006 annual report as my seasonal greeting to my friends. Unlike usual year, I could not deliver this report due to th reason written in [Private], sorry.
Please enjoy it. I hope to keep in touch with you.
Thinking of you from Japan in early 2006
--- My topics in 2005 ---
During the 2005 fiscal year (April, 2005 - March, 2006), I have 1 Bachelor,
5 Master students, 1 Research student, 1 Doctoral student, and maximum
3 postdoctoral researchers and 2 technical staffs.
In the 2006 fiscal year, 2 Master students will join and 2 people will take an enrollment examination of doctoral course in this March to join our lab. Although the exact number of students in 2006 is not fixed at this moment, my laboratory will consist of about 10 students and 3 postdoctoral researchers. Thanks to postdoctoral researchers, they give a good stimulus to students, and we can have research-oriented postdoctoral seminars, while student seminars are for education.
I have worked on the combination of human factor and computational intelligence
that I call as Humanized Computational Intelligence. Interactive evolutionary computation (IEC) is still a main technique
in my research for this direction. I prepare a survey and tutorial paper
of IEC research and a PDF file of slides of IEC applications at the top
of my web page, FYI.
There are two big projects that I am focusing:
They are deeply related each other.
Simple explanation of the latter research would be "to make it clear how to realize more impressive, more calm, and others." When we view movies or listen music, sometimes we become impressive, relax, thrilling, or depressive. These emotional responses are reflected on physiological reactions. The most strong factor giving influence on such emotion responses may be context, e.g. stories of movies, but in the sense level, physical signals must influence physiological responses directly. This research aims to find out physical features that influence physiological responses, find out their relationships, and control the physical features to make human's physiological responses move to the target responses. For the control in the final stage, I am thinking to use Extended IEC based on physiological feedback.
VR Presence vs. VR Sickness is a research of our doctoral student, and its objective is to realize a better artificial (virtual) environment, which is the objective of the 21st Century COE research.
Reduction of IEC User's Fatigue at our lab in 2005 aimed to accelerate EC convergence by learning user evaluation characteristics, making an evaluation model, and predicting user's evaluation, which results to reduce the number of IEC user's evaluations and reduce IEC user fatigue. One of IEC features is that quite few number of generations is allowed due to user fatigue. If this learning user's characteristics takes needs training data in many generations and there is few remained generations in IEC search for using the learnt model, the effect of the proposed learning model is not big.
Here, we introduced a new idea. We embed multiple evaluation models of other several users that are learnt previously into an IEC system and use the models as alternatives of the IEC user until the system leans out the user's evaluation characteristics. The idea is based on "even other users' models are better than nothing." When the embeded models are similar to the user's evaluation characteristics, it must work well; otherwise, it must work worse. We are evaluating this effect through simulation experiments and analyzing the similarity of humans' evaluation for some application tasks. We presented two domestic conference papers.
IEC-based MEMS Design Support System research uses the idea of learning IEC user's evaluation characteristics, too. JSPS Postdoctoral Fellow, Dr. Kamalian, has worked on this topic since he was a doctoral studnet at UC Berkeley. Now, we are testing a neural network (NN) and a fuzzy rule-based system (FL) for the learning models with the cooperation of his colleagues at UC Berkeley.
This support ststem consists of an automated design part using multi-objective optimization and an IEC design part where quantitative evaluation is difficult but can be evaluated based on IEC user's experience and domain knowledge. IEC user fatigue can be decreased by increasing the performance of the auto-design part. Our idea is to introduce a system that predicts the user evaluation to correct the output from the auto-design part and reduce IEC user fatigue. We presented two conference papers in 2005 and am submitting conference papers.
Our research outputs in 2005 was 1 IEC tutorial for AI Encyclopedia, 2 journal papers, 5 international conference papers, 2 tutorial papers for journals, and 6 domestic conference papers. We accepted Alexsandra Britrap, a doctoral course student of Cranfield University in UK, to our laboratory for two months in last summer and conducted a joint research on parallel IEC using her doctoral research task of room-layout design. She compared parallel IEC with conventional multi-objective genetic algorithms and other methods. This research will be presented at European evolutionary conference, EvoWorkshiop2006, held in this April.
I taught courses on computational intelligence, software engineering, data and signal compression, and other lab works in 2005.
Teaching English will be added since April, 2006.