2001 News___________________________________________

Caltech ERC Industry Day

May 2001

The Jisan Research Institute will be participating in the 7th Annual Industry Conference at the California Institute of Technology (Caltech). This event, which brings together business people from many different companies and Caltech scientists, is a chance to let the business community know what is happening at the Caltech ERC, and an opportunity for the business community to have a taste of what's on the horizon for their industries. Many collaborations are born from this event, which is becoming increasingly important in providing for critical lines of communication between industry and science.
The Jisan Research Institute represents another important branch of the Caltech connection between industry and education. While JRI research is largely scientific in nature, many business applications are possible from many of the projects under investigation. In supporting the Jisan Research Institute, the Caltech ERC is helping to make an investment in the future of science in America and of American business. The students coming from the Jisan Research Institute will be sensitized to the needs of science and of industry, and will have eight years of college and graduate school to develop new and interesting ways of applying what they know to both worlds.

April 2001

Jisan Research Institute scientist Dr. Sanza Kazadi and students Yan Qi, Nancy Huang, Isaac Park, Brian Kwan, Paul Hwu, Waynn Lue, and Hubert Li have discovered an evolutionary system that does not benefit from piecewise evolution. The system, which incrementally designs digital adders from single element connectionist networks, is capable of designing complex networks capable of carrying out the computation correctly. Piecewise evolution refers to the design of the system in small incremental steps whereby a single functional unit is created initially, and each additional unit is created correctly before work commences on a next unit. The measure, expected to yield significant advantages in the time to evolution, failed to yield the expected advantages. In fact, there was no significant improvement in the system at all as a result of the staggered nature of the calculation.
This work is important in that it illustrates a counterexample to the notion that breaking up a task will make the task more manageable to an evolutionary algorithm. The key problem lies in the need for each of the system outputs to utilize information from each of the inputs. The complexity of the design derives from each of the individual outputs' complexity rather than the complexity of the integration. Understanding this distinction is an important step in understanding the larger general problem.
A paper entitled Insufficiency of Piecewise Evolution will appear in the upcoming Proceedings of the Third NASA/DoD Workshop on Evolvable Hardware.

New Facility

April 2001

The Jisan Research Institute is moving to a new facility located at 28 North Oak Avenue in Pasadena. This new facility has 2200 square feet and encompasses the entire second floor of the small office building it is house in. The facility will be able to hold a significantly larger computer lab, a well stocked library containing the scientific journals Artificial Life, Evolutionary Computation, and IEEE Transactions on Evolutionary Computation . A robotics laboratory will also be built, allowing both theoretical studies and real studies of swarm engineering at the Institute. Unfortunately, the move will disrupt both email and network traffic during the period of days between April 1, 2001 and April 3, 2001. The fax number will also change as a result to (626) 564-1248.

March 2001

JRI scientists Shaheen Hoque and Dr. Sanza Kazadi, and research students Angel Li, Willie Chen, and Elvio Sadun have completed preliminary work on a new method of image recognition. This work, touted as being a significantly novel approach to a rather well-studied problem in computer science and electrical engineering, is based on the application of nonlinear dynamics to an image recognition system. The basic strategy consists of using the image to construct a nonlinear dynamic system. This system's properties may then be used to build a signature unique to the image.
The system is novel in that it is one of the first that utilize the detailed properties of nonlinear dynamic systems, capitalizing on the manifold structure of the systems. This structure has the desirable property of continuity. This means that images that are similar tend to produce signatures which are similar. This allows one to use one signature or a small region of signatures to signify the image in question. The system has been applied to simple images of rather different shapes, and data has been collected which demonstrates rotational and translational invariance of the system. Current work includes evaluating the system for its possible use as a human face recognition system.
A paper entitled Identification of Shapes Using a Nonlinear Dynamic System detailing this work will appear in an upcoming Lecture Series in Computer Science . A patent in the name of all members of the research team has been submitted based on this work.