Second International Conference on Information Fusion (FUSION99)
submitted to IEEE AES MagazineBy
Peter Willett, University of Connecticut
September 13, 1999Beautiful Sunnyvale California, more specifically the comfortable and surprisingly affordable Sunnyvale Hilton hotel, was the site of this year’s (actually the second) flagship colloqium of the International Society of Information Fusion (ISIF). From July 6-8 some 203 conference attendees – considerably up from last year’s 161 at Fusion'98 in Las Vegas -- from around the world gathered for a companionable and informative stay. Participants were treated to presentations of a total of 187 papers organized in four parallel tracks, as well as 4 plenary talks, a well-catered luncheon and a nicely-lubricated reception.
The key to this conference was diversity. This is partly in terms of affiliation, with slightly less than one half of attendees academic, a third industrial, and the rest from government organizations. And also in background: about half the participants were American, but at least eight came from each of France, Sweden, the UK, Canada, and Japan, with representation from twelve further countries. Perhaps the more interesting diversity, at least from the perspective of one who had until recently thought that data fusion meant using more than one sensor to detect or to track, was in the breadth of approaches to and applications of data fusion.
There were four sessions on target tracking, two on classification, and three others on similar topics. These sessions are the sort I had expected to see, very interesting to me and with titles like "Fusion of Multi-Sensor Information from an Undersea Distributed Field of Sensors" (Hatch, Jahn, and Kaina from SPAWAR), "ARTAS: An IMM-based Multisensor Tracker" (Hogendoorn, Rekkas, and Neven from NAL Netherlands), and "Kernel Based Methods in a Mixture of Experts Framework" (Dodd and Harris from U. Southampton UK). There is also progress in the challenging problem of the fusion of images from different aspects and modalities, with both military ("Testbed for Fusion of Imaging and Non-Imaging Sensor Attributes in Airborne Surveillance Missions", by Jouan, Valin, and Boss, from DRE and Lockheed-Martin Canada) and robotic ("Estimating Two-Arm Distance by Fusion of Distributed Camera-Views", by Scheering, Zhang, and Knoll from Bielefeld, Germany). Data fusion has advanced to the extent that there is now a need for some management of and higher-level thinking about it, a total of five sessions examples of whose papers are "Pitfalls in Data Fusion (and How to Avoid Them)" (Hall and Garga from Penn State) and "An Information System for Object Classification and Situation Analysis using Data from Multiple Data Sources" (Jungert from the Swedish DRE). And there has been some interest in data fusion from the computer science perspective, for example "Semi-Automatic Integration of Knowledge Sources" (Mitra, Wiederhold and Jannink from Stanford) and "Soft-Computing Integrated Circuits for Intelligent Information Processing" (Shibata from U. Tokyo).
Perhaps the original data fusion application was surveillance, but many of the application areas are startlingly novel – again, at least to me. Data fusion has apparently attracted attention in the medical community: an example paper was "Aorta Detection in Ultrasound Medical Image Sequences Using the Hough Transform and Data Fusion" (Debon, Solaiman, Cauvin, Peyronny and Roux, from ENST-Bretagne and U. Brest), in which results from several imaging techniques are fused to give higher resolution of our most important artery. The financial community is also interested: we have, for example, a fun paper "Fusion of Neural Classifiers for Financial Market Prediction" (Keaton from CalTech), in which improved prediction of the US dollar to German mark currency market is achieved by combining several randomly-initialized neural networks. Industry has noticed fusion too, and we have the complementarity of various vibration sensor types and locations applied to fault diagnosis in a diesel engine in "Fault Diagnosis using Multi-Parameter Fusion" (Shen and Tay from Singapore). Finally, we had a perhaps unfocused but definitely exciting collection of presentations in a pair of "Emerging Applications" sessions, examples of these to oil exploration, earthquake monitoring, and freeway traffic management.
Each day of the symposium began with a plenary session. At the first of these, Ren Luo from Taiwan’s National Chung Cheng University gave us an appropriate overview of data fusion: what it is, why we are interested in it, and where is the state of the art. In the second Ken Ford, director of NASA's Center of Excellence in Information Technology in Ames, and Peter Norvig, the Computer Science division head, gave the space agency’s view of data fusion. Artificial intelligence provides a framework in which data fusion naturally and necessarily resides, and NASA is keenly interested for the purpose particularly of remote space exploration; "there is, literally, no other way to make it work". The third day’s plenary was from Franklin White, director of Program Development at US Navy SPAWAR. His perspective was international, and his thesis is that organizations are "drowning in data and information and starved for … understanding" --- thus the importance of data fusion, and the shortsightedness of the fiscal "barbell" of expensive sensors on the input side, expensive actions and capabilities on output, and comparatively meager spending for information extraction and fusion. And for Wednesday’s lunch we had not only a nice meal, but also a plenary talk from Gordon Shaw of UC Irvine’s Center for the Neurobiology of Learning & Memory – thanks to George Chapline of Lawrence Livermore for inviting him. Dr. Shaw’s background is a singular fusion of particle physics and of theoretical neurobiology, and during his fascinating talk he explained, as well as one can to a roomful of engineers, his "Mozart effect". Apparently there is a measurable and even a considerable difference in intellectual effectiveness (e.g. test scores) attributable to hearing the composer’s music. Why this is so is not clear, but there is speculation that the symmetry and patterns in Mozart’s music are able to "prime the pump" for similar thinking tasks. The audience listened to Dr. Shaw’s test music at the beginning of his seminar; unfortunately the Mozart effect is temporary, and we were back to ourselves after lunch.
Credit and thanks are due in a number of directions, particularly to Xiao Rong Li and Mark Bedworth (conference vice-chairs, from U. New Orleans and Jemity in Malvern, the UK, respectively), to Pramod Varshney (program, from Syracuse U.), to Belur Dasarathy (publicity, from Dynetics), to Rob Levinson (publication, from UC Santa Cruz), Chee-Yee Chong (financial, from Booz-Allen & Hamilton), to Fa-long Luo (local arrangements, from ReSound Corp.), and to the conference’s honorary chair Yaakov Bar-Shalom (U Connecticut). Extra special thanks are due to the tireless Capt. Erik Blasch (Wright-Patterson AFB) for doing almost everything to keep the conference running smoothly, and Dongping (Daniel) Zhu (Zaptron Systems) the general chair and the conference’s driving force. Appreciation is also due the conference’s sponsors, the US Army Research Office, NASA Ames Research Center, the IEEE Signal Processing, Control Systems, and (naturally) Aerospace and Electronic Systems societies, and of course the International Society of Information Fusion (Jim Llinas from SUNY Buffalo is the society president).
Please see the conference web site at http://www.inforfusion.org/fusion99/, aggressively maintained by Dan Zhu, and with all its pictures a fun visit. Please also visit the parent ISIF web site for information about next year’s conference - Fusion'2000, not yet available at time of this writing. Fusion99 Proceedings and CD-ROMs can be ordered from the ISIF web site.