Three Computationally Demanding Problems in Search of ASAP Solutions Daniel Lopresti Department of Computer Science and Engineering Lehigh University Bethlehem, PA, 18015, USA lopresti@cse.lehigh.edu Abstract This paper presents an overview of three problem areas where ASAP computing might be profitably applied. Drawn from the fields of pattern recognition, bioinformatics, and biometric security, sufficient detail is provided to explain these selections and the role that ASAP might play, as well as to encourage those so inclined to take a closer look. 1. Introduction Much has changed in the 20 years that have passed since the first conference in this series where I had the good fortune to publish one of my earliest papers [11]. Back then, most of the work I did (including my PhD disserta- tion at Princeton) was aimed at designing and implement- ing fast parallel sequence comparison algorithms for custom VLSI [6, 10] and FPGA [4, 5, 13] architectures. Personal computers were just coming to the fore, and the canoni- cal machine for benchmarking purposes was the DEC VAX 11/780 which was regarded as providing a blazing 1 MIP (million instructions per second). Since that time, processor speeds have increased by three orders of magnitude. Algo- rithms that were much too slow to consider running in 1986 can now be fielded on a pocket-sized PDA. This would seem to spell bad news for application-specific computing. For a variety of reasons not related to the answer to this question, my own research turned to other topics for most of the intervening period. It was with some interest, then, that I began to ponder this point again after spending years away from it. While I hardly expect those active in the field to harbor such doubts, I am happy to report that I was able to (re-)convince myself that the ASAP (Application-specific Systems, Architectures, and Processors) approach is still appropriate and even vital for problems not much different from the ones that first caught my attention long ago. In the area of sequence comparison and text search, it is reassuring to see current work on an implementation of the BLASTN algorithm for a reconfigurable system [9], op- timization of hardware usage for various “flavors” of string matching [3], and a parallel automaton to speed up the Aho- Corasick algorithm for detecting, say, URL and virus pat- terns [21]. While not claiming to be a scientific survey, and un- abashedly influenced by my own personal experiences, this paper presents a view of three problems of much current in- terest where ASAP computing might be profitably applied. These derive from the fields of pattern recognition, bioin- formatics, and biometric security. What are the common features that unite such problems? First, the task should be an important one where at least part of the computation is too slow for realistically-sized instances when run on today’s general-purpose computers. Often this is an inner loop step. Next, it must be possible to bring the problem into the realm of feasibility through the use of an ASAP solution. Finally, when commercial in- terests come into play, the use of ASAP technology must provide enough of a competitive advantage to compensate for the investment of time and money over a simple soft- ware approach. The problems I will present pass the first test above, and it seems likely they will pass the second, although the precise details appear largely unresolved and await the efforts of an enterprising researcher or team. Each problem is explained with enough background to give an interested reader the resources needed to pursue the question further. I suggest why I believe an ASAP approach may be justified, but present no definitive solutions. Indeed, this paper puts forth many more questions than answers. I begin by describing symbolic indirect correlation (SIC), a new approach to pattern recognition when segmen- tation is difficult, but lexical context can be exploited. I then present a computational pipeline for searching the human transcriptome for signs of RNA editing events. Lastly, I dis- cuss the dangers posed to biometric security by generative models for human behavior and the idea of stress-testing such systems. These three topics are drawn from ongoing research projects where I am just one of many collaborators, Application-specific Systems, Architectures and Processors (ASAP'06) 0-7695-2682-9/06 $20.00 © 2006