If this extensive scholarship seems dizzying, Zlatareva, with her amiable smile, makes it comprehensible. AI, she explains, “allows us to write computer programs to solve complex tasks requiring intelligent thinking and problem solving.” It comprises several areas. Games playing, for example, such as the super-computer Deep Blue defeating world chess champion Gary Kasparov.

Expert systems—another AI area—deals with programming computers to make decisions in real-life situations. Some expert systems help medical practitioners diagnose diseases based on symptoms.

An auspicious area of AI is neural networks, which have been proven useful in voice recognition and natural-language processing, allowing people to interact with computers without needing any specialized skills.

Robotics, another AI domain, encompasses programming computers to see, hear, and react to other sensory stimuli, for example, in exploring the oceans.
 

Knowledge Testing

What all these AI applications have in common is knowledge. How reliable is this knowledge? So much depends on the competence of the system, Zlatareva cautions: “If the system processes incomplete or incorrect knowledge, the computer’s advice should not be taken.”

This is where Zlatareva’s scholarship enters the picture. During the 1980s, the professor’s theoretical work revolved around ways to represent knowledge which is imperfect. In the 1990s, she developed techniques and tools for formal verification of computer knowledge, which is called knowledge bases, to prove their correctness, consistency, and completeness.
 

Accomplishments as a Scholar

“One emphasis of my research has been in evaluation and quality assurance of Knowledge-Based Systems (KBS),” says Zlatareva, “to ensure structural and functional properties of a KBS, which is critical for its success as an application product.”

Quality assurance of KBSs is a very broad field, which includes verification of knowledge bases; validation of KBS performance; generation of test cases for system testing; knowledge base refinement, restructuring, updating, and maintenance.

“I believe KBS quality assurance can only be achieved if all these issues are tackled from a common perspective,” elaborates Zlatareva. “My earlier work (before 1999) was concerned primarily with the development of efficient verification techniques for rule-based systems and system testing. However, no matter how good a verification procedure is, the system may still be unacceptable if it is not properly validated and maintained during its entire lifecycle.”

Zlatareva turned her energies to developing an automated refinement procedure for updating and revising system knowledge. During a past sabbatical leave, she says, “I conducted research on application of genetic algorithms in machine learning and theory refinement intended to address this issue. This resulted in a novel technique for restructuring rule sequences capable of improving both the correctness of the knowledge base and the efficiency of KBS performance.”

The research was summarized in a paper presented at the 14th International Florida Artificial Intelligence Research Symposium Conference, 2001, and it appeared in the conference proceedings published by the American Association on Artificial Intelligence (AAAI Press), the most prestigious publisher for AI research.

Another outcome of her research in machine learning was a joint paper with colleagues Dr. Zdravko Markov, CCSU associate professor of computer science, and Ingrid Russell (University of Hartford) on how machine learning can be incorporated into an undergraduate course. Their paper was presented at the 2003 International Conference of Artificial Neural Nets and Neural Information Processing and published in conference proceedings.

Newest Research Challenges

Of late, Zlatareva has been pursuing research evaluating non-monotonic knowledge bases. Generally, the term “non-monotonic” covers inferences in which reasoners draw conclusions tentatively, reserving the right to retract them in light of new information. “We want our computer program to retract its previous conclusions when presented with new facts. If we want to have a realistic, large-scale knowledge base, we have to know how to represent and verify non-monotonic knowledge,” she states. Results of Zlatareva’s research were reported at the 17th International FLAIRS Conference, 2004, and at the 19th International FLAIRS Conference, 2006, and appeared in the conference proceedings.

Zlatareva envisions creating an implementable framework for verification of large knowledge bases, such as those on the Web, to ensure consistency, coherence, non-redundancy, and completeness. “There’s so much artificial intelligence on the Web,” she says, undaunted and eager to learn what she’ll discover.

— Geri Radacsi
 

Back to Courier