Parallel processing systems offer a major improvement in capabilities to AI programmers. However, at the moment, all such systems require the programmer to manage the control of parallelism explicitly, leading to an unfortunate intermixing of knowledge-level and control-level information. Furthermore, parallel processing systems differ radically, making a control regime that is effective in one environment less so in another. We present a means for overcoming these problems within a unifying framework in which 1) Knowledge level information can be expressed effectively 2) Information regarding the control of parallelism can be factored out and 3) Different regimes of parallelism can be efficiently supported without modification of the knowledge-level information. The Protocol of Inference introduced in [Rowley et al., 1987] forms the basis for our approach.
[1]
Walter Hamscher,et al.
Joshua: Uniform Access to Heterogeneous Knowledge Structures, or why Joshing Is Better than Conniving or Planning
,
1987,
AAAI.
[2]
Al Davis,et al.
The Architecture of the FAIM-1 Symbolic Multiprocessing System
,
1985,
IJCAI.
[3]
Michael R. Genesereth,et al.
A Variable Supply Model for Distributing Deductions
,
1985,
IJCAI.
[4]
Salvatore J. Stolfo,et al.
DADO: A Tree-Structured Machine Architecture for Production Systems
,
1982,
AAAI.
[5]
Allen Newell,et al.
Initial Assessment of Architectures for Production Systems
,
1984,
AAAI.
[6]
Charles L. Forgy,et al.
Rete: A Fast Algorithm for the Many Patterns/Many Objects Match Problem
,
1982,
Artif. Intell..
[7]
W. Daniel Hillis,et al.
The connection machine
,
1985
.