One of the earliest overseas partners in RWCP is the Institute of Systems Science, National University of Singapore. Currently, there are two RWC Labs in ISS. One is Neuro-ISS and the other one is Novel Function-ISS.

1. Neuro ISS

The research carried out in Neuro-ISS Lab is under the theme "A New Model of Neural Networks Called Neural Logic Networks". The research team is led by Prof. Hoon-Heng Teh. Other team members are Dr. Ho-chung Lui, Dr. Liya Ding, Dr. Ahhwee Tan, Mr. Joohwee Lim, Mr. Loonin Teow, Prof. Peizhuang Wang, Mr. Tiong-Hwee Goh and Mr. Fon-lin Lai.

The idea of Neural Logic Networks (NLN) model in an attempt to establish a new model as a fusion of neural networks and fuzzy logic. It was motivated by the lack of logical reasoning capability in current neural networks models and the need of learning for fuzzy logic reasoning.

The research of NLN has been carried out on three sub-projects:
(1) Theoretical development and the application to hand writing character recognition.
(2) Neural Prolog -- to develop a flexible knowledge-based inference system where the inference is based on the extended logic and knowledge pieces represented by NLN are used to solve a problem.
(3) Neural Logic Expert System-- to develop expert systems that integrate the complementary strengths of neural network and symbolic AI paradigms. The system can use both of rule-based knowledge and experience-based knowledge for real world decision making.

Combining the research results from early stage, the team has developed an approach of hierarchical interactive Reasoning fOr Information-integration (HIROI).

It enables learning and construction of knowledge from data and employs a unified hierarchy for information integration and reasoning. An application prototype using this approach for document image understanding has been developed.

The same approach can also be applied to various applications where the information from input (human溺machine) is not perfect (the ambiguities and errors may be caused by man-machine interface or mistake by human) but there are knowledge sources which could help to get a better understanding.

We are planing to continue the work on:
(1) the research of different inference strategies for real applications (using soft logics)
(2) the learning and construction based on multi-modal, multi-source knowledge
(3) the research of various models for integration in different levels (for multi-source, multi-expert)
(4) the study on the representation form for knowledge integration (for multi-modal)

2. Novel Functions ISS

The research Theme of Novel Function ISS Lab is "Flexible Storage and Retrieval of Multimedia Information". The principal investigator is Dr. Jian Kang Wu. The members of the research team include: Dr. A. Desai Narasimhalu, Dr. Mohan S. Kankanhalli, Dr. Limsoon Wong, Mr. Chian Prong Lam, Dr. Li Yang, Dr. G. Phanendra Babu, and Mr. Sheng Bin Cui.

The importance of multimedia research has been world-widely recognized. As one of pioneer research groups, we focus on exploration of the theory and realization method for content-based storage and retrieval of multimedia information in a distributed environment.

Breakthrough and progress have been made to the following research topics:

1. Evaluation has become a critical issue for integrated multimedia systems. It provides mechanism to evaluate, test, and benchmark the multimedia database systems. Traditional database benchmarking tests the response time only. In the field of information retrieval (text retrieval) benchmarking has been conducted by evaluating the relevance of the retrieved results by recall and precision. As far as multimedia information concerned, there is no work done on evaluation of content-based retrieval. We have made the first attempt on the evaluation of content-based integrated multimedia systems by proposing a framework as well as definitions and algorithms of the evaluation process. A loss function is defined as a measure to test multimedia information retrieval systems. This measure is quantitative, precise, and covers the concept of recall and precision.

Using loss function as a measure, we have proposed a learning algorithm for similarity function on multi-modal feature measures. It enables the system extensibility and facilitates flexible configuration of the system to various applications. As a result of our exploration, we are able to look one step further into problems of recall, a phenomenon in our daily life. Recognition by recall is different from recognition by classification which is a traditional topic for pattern recognition. The problems there are structured as finding decision boundaries which can partition patterns into a number of disjoint classes.

Recognition by recall views recognition problems from another perspective. It uses similarity rather than discriminant function as the key to solve the problem. You recognize the person since you recall the similar image from your memory. According to this idea, we have successfully recognized the face images of the same person even there are major variations and taken at several age intervals. The data is very rough and is obtained from NIST of USA.

2. As one type of content-based retrieval, we have made significant progress on spatial query language. As rapid development of information highway, more and more application problems will not be solved without spatial data analysis. Spatial query language is a core in the spatial information system. There has been a lot of effort made toward developing a spatial query language. Successful SQL is supported by relation algebra. To develop a useful spatial query language we need to develop spatial algebra as its basis. Based on our previous work on dual data structure, we have proposed and developed a spatial algebra, and the first version of a spatial query language. It can handle queries such as "find me the land where the price is about xx while it is far from other shopping centers", which is believed not to be able to solve by any existing system. It can also provide basic functions for high level decision supporting system.

3. Real world problems are complex. Their solution may involve many data sources which are complex and distributed. Based on our previous work on querying multiple distributed complex databases, we have investigated the extension and new applications of this technique. The results of this research is that we can easily access variety of databases in a large area network, bring the data back, integrate these data and present to the user or feed into another analysis module. The whole process works in a way that the user feels as if he/she is access one database. This system has solved several problems which are claimed not to be able to solve.