|Volume 2: No. 06|
Wavelet theory has drawn the attention of the business community. Some researchers think it's a breakthrough in radar target detection, signal analysis, data compression, medical imaging, and other data analysis tasks. Ronald Coifman, a Yale mathematician, has co-founded Fast Mathematical Algorithms & Hardware Corp. (Hamden, CN) to develop software for decomposing signals into wavelet sets. Aware Inc. (Cambridge, MA) makes wavelet-based chips, underwritten by R&D funds from a $4M DARPA pot; one use is 150:1 compression of digital movies. [John Carey, BW, 2/3.] I haven't followed this area closely. Are wavelet basis sets really better (or faster) than Gabor functions, differences of Gaussians, or binomial waveforms? Regardless, the availability of chips and of government/military/commercial money makes this an interesting research area.
The mirror defect in the Hubble Space Telescope may prove to be a blessing. Although it does prevent certain narrow-aperture experiments, the blur reduces "faint object camera" saturation from very bright sources. Images reconstructed on an MP-1 massively parallel computer (in minutes or hours) are of better quality than with a perfect mirror. This phenomenon is likely to be useful in other imaging domains as well. [Michael Alexander, CW, 1/20.] (Now what we need are holographic detectors.)
Irvine Sensors Corp. (Costa Mesa, CA) is working on an integrated 128 x 128 IR sensing array and neural processor that would pack 128 128-node neural chips into the space of two sugar cubes, for 10^13 interconnections per second -- close to the 10^15 rate of the human brain. Initial funding of $635K is coming from the Strategic Defense Initiative Organization (SDIO) through ONR. [Military & Aerospace Electronics, 11/91.]
Intelligent Computer Systems Research Group (Rockville, MD) has raised money to sponsor the work of others in developing a Learning and Thinking Machine (LTM). Technical analyst Peter Lockwood is a spokesman for ICS Research; other co-founding computer professionals wish to remain anonymous. The group hopes that $5M to $10M over five to eight years will produce a machine able to learn from books, databases, and spoken instruction. One company they regard as promising is Cognitive Modeling Designs (Rockville, MD). [Michael Alexander, CW, 1/20.]
Three researchers at RPI (Troy, NY) have a $300K grant from the Henry Luce foundation to build Autopoeisis 2000, a program to write fiction. Bad fiction, no doubt, but it's a start. The three are David Porush, a literature professor; Marie Meteer, a computational linguist and programmer; and Selmer Bringsjord, an assistant professor of philosophy. [Rick Ratliff, Knight-Ridder. SJM, 2/2.] Sometimes what you can sell is more important than what you can do. The important point, though, is asking the right sponsor. It's very unlikely that NSF would have funded this.