A parallel marker-passing computer for knowledge processing
Year: 2015
Type: dissertation
Abstract: Marker passing is a promising reasoning paradigm for solving many difficult AI problems, including natural language understanding, speech recognition, planning, and knowledge acquisition. Marker passing is inherently parallel and provides a good framework for building a specialized knowledge processing computer.
This thesis is on the Semantic Network Array Processor 2 (SNAP-2), a second generation knowledge processing machine that features many innovative architectural features not found in current computers. Among them are: special knowledge coprocessors, synchronization hardware, and an instruction set tailored toward AI processing. The SNAP-2 architecture is well suited for VLSI implementation. Detailed analysis has been done on the chips required, the interconnection network needed and the implementation strategy.
A register level simulator was built using the Verilog hardware modeling language. Several knowledge processing applications were run and compared with the Connection Machine 2 and the existing SNAP prototype. The simulation results show that SNAP-2 is three orders of magnitude faster than the Connection Machine 2 for a selected set of knowledge processing applications. (Copies available exclusively from Micrographics Department, Doheny Library, USC, Los Angeles, CA 90089-0182.)
Author: Lee, Wing Chung (author)
Language: English
Subfield: Hardware and Architecture
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W2497178299
