ABSTReCT The research activities of the Heuristic Programming Project, for the four-year period ending July 31, 1977, are summarized in this report. Contributions to Knowledge Engineering research in the fields of knowledge acquisition (both interactive and automated), knowledge representation and knowledge utilization were reported in over thirty publications by members of the project. A summary of those publications is?resented here. The Al Handbook, an encyclopedic reference to the field of::tificial Intelligence, is described in the appendix, along with the excecteç table of contents and sample articles.

Over tt7e years a range of different mechanisms have been proposed and used (e.g., standard procedure invocation, goal-directed invocation, etc.), each typically motivated by the attempt to develop new forms of knowledge encoding (e.g., procedures, PLANNER theorems, etc.). We consider in this paper tne strengths and weaknesses of a range of these mechanisms, paying particular attention to their expressiveness and validity. This analysis brings to light certain shortcomings shared to some degree by all current mechanisms. A number of ideas are presented as the basis for a mechanism which appears to offer a way of overcoming the problems discovered. We describe how those ideas have been implemented and tested in a rule-based system, and explore their impact on system performance, ease of construction, and flexibility. We consider also their value as a generalization of the existing notions of procedure calling. Though the terminology may differ, some of the shortcomings we point out and some of the ideas proposed may be recognized by others who have built similar systems, where some of these ideas have been Implemented in various Informal ways. The purpose of this paper is not, therefore, to advocate a particular solution, but instez.d

A range of different encoding techniques have been developed, along with a number of approaches to applying knowledge. Most of the effort to daze however, has concentrated on representing and manipulating knowledge about a specific domain of application, like game-playing ([14D, natural language understanding ([153, [19]), speech understanding ([8], [II)), chemistry ([7]), etc. This paper explores a number of issues involving representation and use of what we term meta-level knowledge, or knowledge about knowledge'. It begins by defining the term, then exploring a few of its varieties and considering the range of capabilities it makes possible. Four specific examples of meta-level knowledge are described, and a demonstration given of their application to a number of problems, including interactive tranfer of expertise and the "intelligent" use of knowledge. Finally, we consider the long term implications of the concept and its likely impact on the design of large programs.

This chapter will discuss one example of how AI techniques are being integrated with, and extending, existing molecular biology sequence analysis methods. AI ideas of complex representations, pattern recognition, search, and machine learning have been applied to the task of inferring and recognizing structural patterns associated with molecular function. We wish to construct such patterns, and to recognize them in unknown molecules, based on information inferred solely from protein primary (amino acid) sequences.

The RNA secondary structure prediction problem (2 RNA) is a critical one in molecular biology. Secondary structure can be determined directly by x-ray diffraction, but this is difficult, slow, and expensive. Moreover, it is currently impossible to crystallize most RNAs. Mathematical models for prediction have therefore been developed and these have led to serial (and some parallel) computer algorithms, but these too are expensive in terms of computation time. The general solution has asymptotic running time exponential in N (i.e., proportional to 2 N), where N is the length of the RNA sequence. Serial approximation algorithms which employ heuristics and make strong assumptions are significantly faster, on the order of N 3 or N 4, but their predictive success rates are low -- often less than 40 percent -- and even these algorithms can run for days when processing very long (thousands of bases) RNA sequences. Neural network algorithms that perform a multiple constraint satisfaction search using a massively parallel network of simple processors may provide accurate and very fast solutions.

How could one represent with a single prototype the class of figures which have an even number of disconnected parts?

How could one represent with a single prototype the class of figures which have an even number of disconnected parts?

Methods of Approach At the outset it might be well to distinguish sharply between two general approaches to the problem of machine learning.

The editors have provided him with special introductory sections as a ing guide to a thorough understand of the reports.

Empirical rule learning and analytic Most learning is based on experience, and this requires a learning methods have predominantly used the first path, representation for the experiential input given to the whereas connectionist systems have relied on the second.