Marshall I. Schor

A continuous real-time expert system for computer operations

The Yorktown Expert System/MVS Manager (or YES/MVS for short) is a continuous real-time expert system that exerts active control over a computing system and provides advice to computer operators. YES/MVS provides advice on routine operations and detects, diagnoses, and responds to problems in the computer operators domain. This paper discusses the YES/MVS system, its domain of application, and issues that arise in the design and development of an expert system that runs continuously in real time.

Making Watson fast

IBM Watson™ is a system created to demonstrate DeepQA technology by competing against human champions in a question-answering game designed for people. The DeepQA architecture was designed to be massively parallel, with an expectation that low latency response times could be achieved by doing parallel computation on many computers. This paper describes how a large set of deep natural-language processing programs were integrated into a single application, scaled out across thousands of central processing unit cores, and optimized to run fast enough to compete in live Jeopardy!™ games.

Match algorithms for generalized Rete networks

Abstract We present a match algorithm which operates correctly on generalized Rete networks that allow arbitrary association of patterns. The original OPS5 Rete algorithm implements matching for left-associative joins; later work generalized this algorithm to support arbitrary join association. However, the naive extension of the OPS5 Rete algorithm to support arbitrary join associations has some subtle flaws (sometimes not all matches are produced with no duplicates) when processing networks having reconvergent (or self-joined) nodes. As a pedagogical means, we first describe an abstract match algorithm for an abstract Rete network having additional edge memories, which provides a useful model for general Rete operations. We then develop two practical specializations of the abstract algorithm, and prove their correctness. The first specialization corresponds to traditional Rete match implementation techniques that are guaranteed to operate correctly for generalized Rete networks. The second specialization is formulated in a general way that allows for delaying match processing at arbitrary stop points within the Rete, with correct match results available on demand at a later time. One current system uses a derivative of this algorithm to support Rete-based matching on demand, complementing the normal match computations done on data-change.

Dynamic augmentation of generalized Rete networks for demand-driven matching and rule updating

Algorithms for dynamically augmenting an existing Rete network having nonempty partial match memories are described. Two uses of this operation are presented: (1) adding a new pattern to the Rete network where it both shares match nodes with the existing Rete network and matches existing data, and (2) implementing an efficient method of demand-driven pattern matching which makes it possible to match a pattern against existing data when demanded, while keeping that pattern out of the normal Rete pattern matching. The algorithms described are applicable to generalized Rete networks in which arbitrary pattern association is allowed. The tradeoffs involved in the design are discussed, including considerations for compilation.<<ETX>>