Signal-to-Symbol Transformation: HASP/SIAP Case Study

level of the hypothesis is referred to as an expectation-lank. The node at the end of an expectation-link is a model-based hypothesis element, and the link represents support from above (i e , the reason for proposing the hypothesis element is to be found at the higher level). 2. ,4 link which goes in the opposite direction, from lower levels of abstraction to higher, is referred to as a reduction-/ink The node at the end of a reduction-link is a data-based hypothesis element, and the link represents support from below (i e , the reason for proposing the hypothesis element is to be found at a lower level). An example of hypothesis elements generated by the KSs are shown in Figure 4 Knowledge about how to use other knowledge: Since the strategy is opportunistic, the program must know when an opportunity for further interpretation has arisen and how best to capitalize on the situation. In HASP/SIAp this type of knowledge is made explicit as will be explained in the following section. How the knowledge is organized and used. How well an Expert System performs depends both on the competence of the KSs and on the appropriate use of these KSs Thus, the primary considerations in the design of Expert Systems revolve around the availability and the quality of the KSs, and the optimal utilization of these KSs When and how a KS is used depends on its quality and its relevancy at any given time The relevance of a KS depends on the state of the CBH. The control mechanism for KS selection needs to be sensitive to, and be able to adjust to, the numerous possible solution states which arise during interpretation. Given this viewpoint, what is commonly called a “control strategy” can be viewed as another type of domain-dependent knowledge, albeit a high level one. Organizing the knowledge sources in a hierarchy is an attempt to unify the representation of diverse knowledge needed for the interpretation task In a hierarchzcally organized control structure, problem solving activities decompose into a hierarchy of knowledge needed to solve problems. On the lowest level is a set of knowledge sources whose charter is to put inferences on the CBH We refer to KSs on this level as Speczalists. At the next level there are KS-activators that know when to use the various Specialists. On the highest level a StrategyKS analyzes the current solution stat’e to determine what information to analyze next. The last activity is also known as focussing-of-attention Kinds of knowledge represented. There are several kinds of knowledge used in HASP/SUP, each represented in a form that seems the most appropriate. Knowledge about the environment: The program must know about common shipping lanes, location of arrays and their relation to map coordinates, and known maneuver areas. This knowledge is represented in procedures that. compute the necessary information. Knowledge about vessels: All the known characteristics about vessel types, component parts and their acoustic signatures, range of speed, home base, etc , are represented in frame-like structures. These constitute the static knowledge used by rules whenever a specific class of vessels is being analyzed. In addition, when some vessel class is inferred from a piece of data, detailed information is available to help make that hypothesis more credible. The use of this information by model-driven KSs reduces the amount of computation by directing other KSs to look for specific data Merpretation knowledge: All heuristic knowledge about transforming information on one level of the CBH to another level is represented as sets of production rules The rules in a KS usually generate inferences between adjacent levels However, some of the most powerful KSs generate inferences spanning several levels For example, a line with particular characteristics may immediately suggest a vessel class. This type of knowledge is very situation-specific. It was elicited from human experts who know and use much of the specialized, detailed knowledge now in the program (It is said that chess masters can immediately recognize approximately fifty thousand board patterns.) There are more examples of rules in the next section. The execution cycle consists of 1 focussing attention on pending time-dependent activities, on verification of a hypothesis, or on one of the hypothesized elements; 2 selecting the appropriate KSs for the attended event;