Microprocessing and Microprogramming 38 (1993) 801-802 North-Holland
Session H2:
801
Decision Mni~ing Under Uncertainty
Chairman:
Alan Rea Parallel Computer Centre The Queen's University of Belfast U.I~
The papers presented in this session address the difficult problems encountered when trying to analyse an uncertain and extremely variable situation. Douglas Adams possibly struck the right note when he wrote "we demand rigidly defined areas of doubt and uncertainty". The designers of many computer systems are striving to attain a clear definition or means of defining a state which is infinitely variable. This goal has been tackled in a variety of ways. One technique, the neural network approach, attempts to simulate the complex decision making processes of the h u m a n brain. Yet other researchers adopt modelling methods which tend to be problem oriented solutions. Using the most complex analytical engine available, the h u m a n brain, even the most mundane decisions are taken on the basis of some element of uncertainty. Thus the need to resolve the uncertain is a critical element when trying to develop computer systems which are required to mimic some element of the h u m a n capacity for problem solving or decision making. In the first paper entitled "Pen Based Recognizing of Hand-Printed Characters" by Klauer and Waldschmidt of the J.W. GoetheUniversity, Frankfurt, their approach to object recognition, namely hand written text, is described. This research project has been implemented on an INFOS Notepad 386 SX pen computer and an associative processor the AM 3. There are two interesting elements
to this paper, one is the description of the character recognition algorithm and the other the AM 3 implementation. The paper provides an excellent overview of the algorithm in question, which has been written in C ÷+ with a view to portability but this has then been optimised for maximum performance on the chosen 386 based machine. The character recognition is a two phase operation following the preparation of the input data by a pre-processor. First, a pattern matching phase is employed to compare the character with a database of known patterns. If this yields an ambiguous result, for example 8 or B, then the second phase resolves these by analysing the formation of the character in terms of the number and nature of the strokes used in its construction. The second implementation which was targetted at the AM 3 seeks to exploit the associative parallelism within this architecture. This is not covered in depth and may, hopefully, be the subject of a future paper. The authors express a desire to use the the pattern matching algorithm to develop a speech recognition system which would also exploit the AM3's associative parallelism. The second paper presented in this session is by Goseva-Popstojanova and Grnarov, from the University "Kiril i Metodij" in Macedonia, and is titled "N Version Programming with Majority Voting Decision: A Dependability
802
Session H2: Decision making under uncertainty
Modeling and Evaluation ~ . This paper describes the authors' work in the area of the derivation of closed form solutions to the failure probabilities of the N version programming system which is subject to coincident failures. The work is based on a two dimensional, continuous time, Markov model which is analysed in the time domain.
In this way the authors account for the interactions between faults in the versions and faults in the majority voter. Also discussed in the paper is the relationship between two of the key features of a faulttolerant, real-time system namely, safety and reliability.