Massively parallel symbolic computing
Advances in hardware and computer architecture continue to change the economics of various AI (as well as all other) computing paradigms. The new generation of massively parallel machines extends the potential for applications at the high end of the computing spectrum, offering higher computing and I/O performance, much larger memories, and MIMD as well as SIMD capabilities. Computing costs for the same level of performance are substantially less, and will continue to drop steeply for the foreseeable future.
All this has clear consequences for AI: for example, larger knowledge bases can be stored; hand coding will continue to become less cost-effective relative to learning and simple-to-program brute-force methods as time goes on; and just about any parallel AI paradigm should be capable of executing efficiently.
A brief overview will be provided of recent successful Connection Machine projects: automatic keyword assignment for news articles using MBR nearest-neighbor methods (MBR = Memory-Based Reasoning); automatic classification of Census Bureau returns; protein structure prediction using MBR together with backpropagation nets, and statistics; work on “database mining”; and Karl Sims' generation of graphics using genetically-inspired operations on s-expressions.
Unable to display preview. Download preview PDF.
- 1.Appel, M. & E. Hellerman. “Census Bureau Experiments with Automated Industry and Occupation Coding.” Proc. Amer. Statistical Assoc., 1983, 32–40.Google Scholar
- 2.Creecy, R., B. Masand, S. Smith, & D. L. Waltz. “Trading MIPS and Memory for Knowledge Engineering.” CACM 35, 8, August 1992, 48–64.Google Scholar
- 3.Goldberg, D. E. Genetic Algorithms in Search, Optimization & Machine Learning. Reading, MA: Addison-Wesley, 1989.Google Scholar
- 5.Koza, J. Genetic Programmming: On the Programming of Computers by Means on Natural Selection. Cambridge: MIT Press, 1992.Google Scholar
- 6.Masand, B. “Effects of query and database sizes on classification of news stories using memory based reasoning.” AAAI Spring Symposium on CBR, Stanford, March 1993.Google Scholar
- 7.Masand, B. “Optimizing Confidence of Text Classification by Evolution of Symbolic Expressions.” Unpublished paper, Thinking Machines Corp., Cambridge, MA, 1993.Google Scholar
- 8.Masand, B., G. Linoff, & D. L. Waltz. “Classifying news stories using memorybased reasoning.” Proc. SIGIR Conf., Copenhagen, July 1992.Google Scholar
- 10.Quinlan, R. “Learning efficient classification procedures and their application to chess end games.” In R.S. Michalski, J. Carbonell, & T. Mitchell (eds.) Machine Learning: An Artificial Intelligence Approach, Los Angeles: Tioga Publishing, 1988, 463–482.Google Scholar
- 11.Rumelhart, D, J. McClelland, et al. Parallel Distributed Processing, Cambridge: MIT Press, 1986.Google Scholar
- 13.Stanfill, C. & D. L. Waltz. “Toward Memory-Based Reasoning.” CACM 29, December 1986, 1213–1228.Google Scholar
- 14.Waltz, D. L. “Memory-Based Reasoning.” In M. Arbib & A. Robinson (eds.) Natural and Artificial Parallel Computation, Cambridge: MIT Press, 1989, 251–276.Google Scholar
- 15.Waltz, D. L. “Massively Parallel AI.” Proc. AAAI-90, Boston, 1117–1122.Google Scholar
- 16.Waltz, D. & J. Feldman, Connectionist Models and Their Implications, Hillsdale, NJ: Ablex Publishing, 1988.Google Scholar
- 17.Zhang, X. & J. Hutchinson. “Practical Issues in Nonlinear Time Series Prediction.” In A. Weigend & N. Gershenfeld (eds.), Predicting the Future and Understanding the Past: Proceedings of the 1992 Santa Fe Institute Time Series Competition, Addison-Wesley, 1993, to appear.Google Scholar