Speaker: Prof Norman Foo

Abstract:  An emerging area in the evaluations of agent behavior, e.g. the ranks in eBay, is the theoretical basis of agent reputation computed from the evaluations.  The talk will survey the area, highlighting key issues, and then focus on one of several approaches.

Speaker: Prof. Dr. Horst Bunke

Abstract:The most common representation format of objects in intelligent information processing is feature vectors. Using this format, an object is modelled by a point in an n-dimensional feature space. Such a representation offers a number of advantages. For example, a large number of algorithmic tools for tasks such as classification, clustering, or dimensionality reduction are available. Moreover, the computational complexity of these methods is usually quite moderate. However, object representation by means of feature vectors also implies some shortcomings. For example, it is not possible to explicitly represent any relations that may exit between various parts of an object. Furthermore, one is confined to always using the same number of features regardless of the size and complexity of the individual objects under consideration. For these reasons, graphs have emerged as an interesting alternative to feature vector based object representation recently. In a graph representation one can use an arbitrary number of nodes and edges to model entities of the underlying problem domain. Moreover, various relations (e.g. spatial, temporal, conceptual, and others) can be easily captured by a graph. As a matter of fact, graphs have gained increasing popularity in various application domains, for example, bioinformatics, computational chemistry, and pattern recognition. However, using graphs for object representation, one also faces some problems. First the computational complexity of many operations is significantly increased when compared to the corresponding operations in a vector space. Secondly, there is a severe lack of algorithmic tools for graph, which is due to the fact that many mathematical operations (e.g. computing the sum, the product, or the mean) are not available in the domain of graphs.

In this talk we first give a general introduction to the use of graphs for formal object representation. Then we review recent work that aims at transferring established methods. Finally, we review some recent work to address the problem of increased computational complexity.   

Speaker: Dr. Sheng-Chuan Wu

Abstract: Life science is characterized by its diversity of information domains, differences in taxonomies from different disciplines, enormous complexity and sizes, and, most of all, intrinsic uncertainty and very fast evolution of knowledge space.  Considering just one biomedical taxonomy (ontology) from NCI, it contains more than 1 million concepts and more than 5 millions of relationships between these concepts, and this taxonomy continues to evolve rapidly as we gain more inside into biological and medical sciences.  How to represent this intertwined information in a consistent and integrated form is an information modeling problem of mammoth proportion.  Try to abstract patterns or extract knowledge from this gigantic information repository is worse than finding a needle in a hay stack.

In this talk, we will show how to use the current RDF semantic database to represent information in life science, use standard pattern matching languages to extract knowledge or apply graph analysis algorithms (e.g., social network analyses) to find the needle in the life science hay stack.

Speaker: Prof. Eric Tsui

Abstract: Taxonomy underpins many of the website and directory navigation schemes for content retrieval. However, information or content navigation support through taxonomy is often constrained due to its inability to take into account the full nomenclature and cultural nuances of information seekers. The emergence and increasing acceptance of social bookmarking and content tagging tools have given birth to lightweight and informal conceptual structures called folksonomies which, as illustrated in this paper, can be used to enhance taxonomy navigation, information search and retrieval. In this paper, we present a novel algorithm for integrating folksonomy with taxonomy through unsupervised data mining techniques, namely Formal Concept Analysis (FCA), K-Means and Simple Matching Coefficients (SMC). The end product is a hybrid taxonomy-folksonomy. Preliminary results have shown that the adopted techniques are promising and feasible as they can be used to integrate a filtered folksonomy with a pre-defined taxonomy for enhancing knowledge navigation. Considering the disparate information classification schemes that often exist in organizations and the abundance of folksonomy and tags, the outcome of this research is of significant industrial relevance with far reaching implications.

Speaker: Prof John Sowa

Abstract: During the past half century, the field of artificial intelligence has developed a large number of theories, paradigms, technologies, and tools.  Many AI systems are based on one dominant paradigm with a few subsidiary modules for handling exceptions or special cases.  Some systems are built from components that perform different tasks, but each component is based on a single paradigm.  Since people freely switch from one method of thinking or reasoning to another, some cognitive scientists believe that the ability to integrate multiple methods of reasoning is key to human-like flexibility.

In his book “The Society of Mind”, Marvin Minsky presented an architecture for intelligence based on a society of heterogeneous agents that use different reasoning methods to solve different problems or different aspects of the same problem.  That idea is intriguing, but it raises many serious issues:  how to coordinate multiple agents, distribute tasks among them, evaluate their results, encourage agents that consistently produce good results, inhibit agents that produce misleading, irrelevant, or unfruitful results, and integrate all the results into a coherent response.  The most difficult problem is to enable multiple heterogeneous agents, acting independently, to produce the effect of a single mind with a unified personality that can pursue globally consistent goals.  This talk discusses ways of organizing a society of heterogeneous agents as an integrated system with flexible methods of reasoning, learning, and language processing.

Speaker: Prof Abdul Sattar

Abstract: Artificial Intelligence (AI) emerged as a new field of science and engineering about 6 decades ago. Since then concerted efforts have been made on designing and developing expressively adequate languages to represent knowledge about real world domains, and building computational tools to efficiently reason with these representations. Indeed, these two aspects of intelligent systems remain the fundamental challenges of AI.  This talk will first give an overview of what has been achieved so far to address these basic problems. We will then focus on the constraint satisfaction paradigm that has become a realistic approach to model real world problems and solve them efficiently using general purpose constraint solving techniques. We will present some of our recent successes on solving the propositional satisfiability challenges, and some open issues.  The talk will conclude with a briefintroduction of the Advanced Technologies for Optimisation and Modelling In Constraints (ATOMIC) project being carried out at National ICT Australia (NICTA).

Speaker: Prof Henry Lieberman

Abstract: Traditional knowledge management is focused on representing knowledge that is special in some way: unique to a person or group; technical or specialized knowledge; specific situation-dependent data about people, things or events. What everybody forgets is that that specialized knowledge builds on a base of Commonsense knowledge -- simple, shared knowledge about everyday life activities. A database might represent an airline flight with airline name, flight number, origin and destination time and place, etc. But no database represents the fact that if you traveling less than a kilometer, you can walk; if you are traveling thousands of kilometers, you probably need to fly.

Why bother to represent this obvious knowledge explicitly, since everybody knows these things already? Because computers don't. If we would like to have computers be helpful to people, avoid stupid mistakes, and make reasonable default guesses about what people might want, they have to have Commonsense knowledge. I will present Open Mind Common Sense, a project to collect human Commonsense knowledge; ConceptNet, its semantic representation; and AnalogySpace, a new reasoning technique that draws plausible inferences, despite the fact that our knowledge base is incomplete, imprecise, and inconsistent.