New contract assessment-Calculates contract conditions when a new contract is made, based on status of various illnesses. Health insurance assessmentSupports judgement of amount paid for hospitalization and operations. Plant fire/explosion risk diagnosis-User enters company and plant name, etc., and risk of fire or explosion is predicted. Fire prevention diagnosis report generation-Possible risks, foreseeable dangers and improvements are displayed to facilitate creation of fire prevention diagnosis reports. Finance Budget assessment-Budget requests from branches for equipment are checked. Retirement consultation-Calculation of retirement allowance is performed as well as consultation on early payment of retirement. Optimum plans for land useUser enters location and surrounding environment. High-return money management plans-User enters return, convertibility and safety issues, system creates asset management plans. Asset management consultation -Interactive input of client's family and assets is performed. These data are used to make advice on asset management. Credit check-Checks applications for credit cards for eligibility. from daily maintenance, system judges cause of trouble and instructs repair personnel on proper measures. Electronic data processing design support-Determines configuration of hardware and software and automatically generates system parameters from estimates of memory and disk capacity. (Examples of detailed descriptions are available.) Finance • Budget Assessment Expert The Dai-Ichi Kangyo • Design Support System for The Dai-Ichi Kangyo Bank, • Expert System for Business Hitachi Credit Corporation • Expert System for Inheritance Advisory Service The Nishi-Nippon Bank, • Development Forecast Expert System The San-in Godo Bank, • Pension Consulting Expert The Sanwa Bank, Ltd. • Private Placement Straight Bond Advice System The Sanwa Bank, Ltd. Insurance Expert Medical Underwriting System Daido Mutual Life Insurance Company Manufacturing • Defect Diagnosis Expert System for Color Picture Tube Manufacturing Fax: (0492) 96-6005 HITACHI.CO.JP or MOTODA@VM.USC.EDU Dr. Motoda was my host on a visit to Hitachi's Advanced Research Lab (ARL), and I reported at that time ["hitachi", 12 Dec 1990] about ARL. Several of the same projects were also presented to JTEC, although at a much later stage of research. There were some new activities too. As readers will discover by perusing the list below, research here is very high class and at an advanced state. Automatic Knowledge Reformation-Basically this research's goal is to provide tools to make knowledge base-building easier. Specifically, the plan is to automate the generation of an hierarchical, multilevel representation of complex data. The major activity has been to implement a program called Concept Learning from Inference Pattern (CLIP). The major application area is the electric circuit domain. CLIP analyzes the inference process about the behavior of the physical device, and extracts the repetition of the typical inference pattern (i.e., the typical behavior of the device). It generates new abstract level descriptions by replacing the set of descriptions relating to a typical behavior with a single new description. The algorithm is similar to a genetic algorithm in order to avoid exhaustive search. CLIP can restructure a single knowledge base into a more efficient form. It is claimed that inference costs are reduced to one half, or in some cases to less than one tenth. A major remaining research task is to study how to use multiple knowledge bases. (In the circuit case, these could be mechanical, thermodynamical, electromagnetic, etc.) Another research issue is to unify the induction from data with CLIP. I was also told that a second major application domain is solving first order ordinary differential equations, however I have had no opportunity to see this in action. Discovering characteristic chunks -This research is based on the observation that when we reason, we always think of some "imagistic" chunks relevant to the reasoning process. (Imagistic reasoning is sometimes called visual reasoning.) The goal is to study how and in what situations these imagistic chunks should be acquired and used, thereby hoping to realize sophisticated reasoning in a machine. The domain area is plane geometry. The idea is that experts can recognize a whole figure as a composition of several characteristic figure chunks, giving characteristic figure chunks, giving them a perspective view. The set of chunks will be stored in the knowledge base to be used in solving future problems. The proposed learning theory can be applied to derive these chunks out of the problemsolving process. Plans are to collect a set of these chunks and evaluate quantitatively how much improvement in the reasoning can be gained by using them in solving unseen problems in the future. A related research area is Explanation-based learning (EBL). Hitachi's idea is to use a new concept to address this problem. They call this recognizability, see Masaki Suwa and Hiroshi Motoda, "Learning Abductive Strategies from an Example", Ohio-State Univ Tech Report, 91-JJ-Workshop. Development of a massive memory workstation-The goal is to explore the architecture necessary to develop a personal computer with gigabytes of memory. It seems that large memory requirements are common in AI applications, and clearly could be of great use in database, CAD, and language processing, among other fields. A prototype workstation has been developed with 64MBytes of memory, using an R3000 (MIPS) CPU, and an Ethernet interface. Current work involves porting Unix as an operating system. There is also work building a nonvolatile read/write 4GB memory machine using an R4000 (MIPS), this also has an Ethernet interface. It is expected that one to two years are required for it to start running. Research plans involve development of a single level storage (integrated memory and file space), precise memory protection (200-500KByte object-wise), and persistency support (incremental execution). In the long run, the plans are to study a new programming model, since it is claimed that current models focus on minimizing code and data size at the expense of execution speed. There is no question that if such a model could be developed to take advantage of huge memory space, it could have a very significant impact on algorithmic development. With memory prices dropping, especially for DRAM, this is an important research direction to follow. Motoda is thinking of the future, when a workbench for intellectual activities will require a terabyte of main memory, and a few tens of gigabytes of memory will be widely available within a few years. Information processing architecture-The goal is to establish a new framework for an architecture suited to simulate the recognition ability of humans (observation, identification, decision, action, experience, and learning). This is basically theoretical research, but the plans are to experiment with current architectures and extract their problems. The experiments are being done through a small vehicle with information gathering and decision capability in a real environment. The vehicle has two video cameras that feed to an image processing board on a Sun. Data are processed by a software tool kit that then connects with a control data transducer and is eventually transmitted to the servo motors of the vehicle. The software is built on Motif in object oriented style. The researchers comment that other labs such as MIT, CMU, and Tsukuba have also been working on vehicles for uncontrolled environments. But their main stress has been on the vehicles, whereas Hitachi's research is mainly directed toward a new architecture (hardware and software) suited for mechanization of higher level recognition, using the vehicles only as experimental tools. Logical interpretation of English sentences-The goal is to perform natural language understanding through the automatic translation of English sentences into logical form. Researchers have made precise the quantification of indefinite noun phrases, and are working on ellispis, disambiguation, anaphora, plural expressions, etc. It is claimed that their work is an improvement over the Core-Language-Engine developed at SRI. Formal theory of discourse understanding-The goal is to construct formal (logical) characterization of discourse semantics. The long term goal is to implement ideas gained here on machine translation or user interface systems, but research so far has concentrated on problems of metonymy. Discourse processing is an active research area, with papers from universities (Stuttgart, Edinburgh, Stanford) as well as government and industry (U.S. Naval Ocean Systems Center and SRI in the United States, and ETL, NTT, NEC, Fujitsu, and Matsushita in Japan). Formalized programming methods-Researchers have proposed a new programming model (set theoretic programming) and formulated a systematic derivation method of algorithms based on structural induction. Most (not all) textbook algorithms can be derived in this way. Current research is directed toward making this more powerful by using combinations of optimization patterns. Work on formalized programming goes back to E. Dijkstra and B. Nordstrom. Domain theoretic study of data types with equations-The goals are to develop more expressible pro gramming/specification languages and new semantic theories with more analyzing power. One method, denotational semantics, is a theory based on domain concepts and has been recognized as very useful. The algebraic specification based on equational theories has been thought of as a promising method for formal specification of software. Current research is to clarify the domain theoretic aspects of equationally defined data types. Work in this area is at a very early stage. There is related work at CMU in the United States, Edinburgh in Scotland, and at Pisa and Torino in Italy. In addition to these projects we were also told that there are activities in Case Based Reasoning (Retrieval of reusable cases) Large Scale Knowledge Integration (Text based knowledge A New Fast Inference Algorithm (CREST, Conflict Resolution Strategy oriented matching algorithm) THE HONG KONG UNIVERSITY OF SCIENCE AND TECHNOLOGY AND THE CHINESE UNIVERSITY OF HONG KONG Presented is an update on progress at Hong Kong's newest university, Hong Kong University of Science and Technology, and an introduction to the Chinese University of Hong Kong. by David K. Kahaner THE HONG KONG UNIVERSITY OF SCIENCE AND TECHNOLOGY Science Hong Kong University Tel: +852 358-7010 HKUS&T was incorporated in 1988 and entered its first class of 560 undergraduates and 140 graduate students in October 1991. There are three schools (Science, Engineering, Business & Management) that provide full doctoral programs. Another school (Humanities & Social Science) provides service courses to the undergraduates; it also has its own graduate program. In addition, there are several interdisciplinary Research Institutes and a Technology Transfer Center. Motivation for a new university came from a sense shared by community leaders that Hong Kong's economic base has been veering away from cheap-labor-dependent industries to more technological industries requiring skilled workers. Cheap labor was a traditional source of Hong Kong's growth, but now manufacturers are moving "off shore." This has led to rapid economic growth by taking advantage of cheap labor in China. A large fraction of Hong Kong's economic growth is currently built on "re-exports," from China on to the West, with rapid increases to Germany and France. Hong Kong's chief industries are still the production of clothing, metals and metal products, textiles, and electronic and plastic products. Currently, Hong Kong manufacturers employ about 2.5 million people (some estimates are as high as 4 million), but only 750,000 are in Hong Kong. It was realized that the future would require even more bulk manufacturing in China and that Hong Kong needed to provide more of the intellectual product. At the same time, Hong Kong has priced itself out of the competitive pool of traditional manufacturing processes, and it was felt that Hong Kong has not kept pace with technological advancements. Other countries in the region spend far more of their gross domestic product on research; Taiwan and Singapore spend 1.5% to 2.5%; South Korea spends 3%; and Hong Kong spends 0.05%. Earlier |