mwang@watmath.UUCP (mwang) (07/05/84)
_D_E_P_A_R_T_M_E_N_T _O_F _C_O_M_P_U_T_E_R _S_C_I_E_N_C_E _U_N_I_V_E_R_S_I_T_Y _O_F _W_A_T_E_R_L_O_O _S_E_M_I_N_A_R _A_C_T_I_V_I_T_I_E_S _A_R_T_I_F_I_C_I_A_L _I_N_T_E_L_L_I_G_E_N_C_E _S_E_M_I_N_A_R - Monday, July 9, 1984. Dr. John Tsotsos of the University of Toronto will speak on ``Knowledge Organization: Its Role in Representation and Decision-Making for Expert Sys- tems.'' TIME: 3:30 PM ROOM: MC 5158 ABSTRACT The so-called ``first generation'' expert systems were rule-based and offered a successful framework for building applications systems for certain kinds of tasks. Spatial, temporal and causal reasoning, the structuring of knowledge, and the explanation of knowledge in terms of knowledge abstractions and com- parisons are among the topics of research for ``second generation'' expert systems. It is proposed that one of the keys for such research is knowledge organiza- tion. Knowledge organization determines control struc- ture design, explanation and evaluation capabilities for the resultant knowledge base, and has strong influ- ence on system performance. We are exploring a frame- work for expert system design that focuses on knowledge organization. In particular, the representation facil- itates and enforces the semantics of the organization of knowledge classes along the relationships of generalization/specialization, decomposition/aggregation, temporal precedence, causal dependency, instantiation, and expectation-activated similarity. A hypothesize and test control structure is driven by the class organizational principles, and includes several interacting dimensions of search (data-driven, two forms of hypothesis-driven, temporal, causal, and failure-driven search). The hypothesis ranking scheme is based on temporal cooperative compu- tation with hypothesis ``fields of influence'' being defined by the hypothesis' organizational relation- ships. This control structure has proven to be robust July 5, 1984 - 2 - enough to handle a variety of interpretation tasks for continuous temporal data. In addition, explanation capabilities are enhanced by the multiple organization principles of the knowledge classes as well as by the resultant inherent redundancy. A variety of queries involving different forms of knowledge abstractions and comparisons can be handled. This framework has result- ed in a representational language called PSN (Procedur- al Semantic Networks), and two medical expert systems, ALVEN (for the assessment of left ventricular perfor- mance from X-ray films) and CAA (for the analysis of arrhythmias). July 5, 1984