Richard Belew - US grants

The goals of this project, called "Academic Institutional Memory" (AIM), are to: refine a set of tools that encourage researchers to increase their use of electronic information sources during research;incorporate machine learning mechanisms that are capable of transforming these researchers' browsing behaviors into self-organizing information structures; and analyze the information structures built manually by the researchers and automatically by the learning mechanism as "artifacts", created by the cultural process of science. In addition to its contribution to machine learning and the philosophy of science, this kind of theoretical analysis is necessary to understand how new information retrieval tools can become part of the infra-structure coordinating the activities of modern scientists. The work under this award will revolve around three related themes: the development of a theory of science and scientific activity as a cultural process; implementation of a prototype of the AIM system as a data collection technology that will permit the testing of hypotheses from the theory; and development of computational models that capture the regularities predicted by the theory.

9221276 Cottrell Learning Semantic Representations for Information Retrieval This is the first year funding of a three-year continuing award. The objective of this project is to develop methods to automatically represent text-based documents from a large collection in a way which facilitates semantically precise retrieval. A critical problem in representing documents is that words in the documents are not accurate descriptors of document content. This is in part due to the polysemy of natural language: A single concept can be described in many different ways. Most current approaches fail to account for this, as they determine semantic relevance using co-occurrence of words in documents. The approach is to index documents so that they are representationally similar when they are semantically related, not just when they coincidentally share terms. Multidimensional Scaling (MDS) and Neural Network theory are foundations of the work. This approach is demonstrated to be similar to the best current technique for statistical semantic analysis of documents: Latent Semantic Indexing (LSI). The work suggests a generalization of LSI, a linear and metric technique, to non-linear and non-metric techniques. This work is expected to provide a well-founded theoretical framework for document indexing based on MDS, to advance the use of neural network techniques in document indexing, and to help in the quantitative evaluation of current document retrieval methods. ***

EIA-98-02219
Joseph Pasquale
University of California, San Diego

CISE Research Infrastructure: The UCSD Active Web

UCSD is investigating system and application support issues for a next-generation World Wide Web, called the "Active Web." The Active Web is premised on the support for active content, content that is rich in multimedia and references to other objects, and for mobile agents, programs that can move about and execute on remote servers, carrying out requests at a distance on behalf of clients. These servers are no longer passive databases as in today's Web, but context-sensitive knowledge networks that contain all kinds of active content. Between the servers, there is a constant exchange of agents, which add to, refine, form interconnections, and make consistent, the distributed content. In the Active Web, there is a high degree of resource sharing, usage is bought and sold as in a market economy, and security is paramount. This grant will allow UCSD to purchase large-scale computer and storage servers and a high-speed network that will connect the various laboratories, and will form a small-scale Active Web prototype.

The project is taking a department-wide coordinated approach, integrating the research efforts in systems, security, multimedia, content-based search, scientific metacomputing, and, tools for software/hardware design and analysis.

Stochastic and population-based search methods (simulated annealing, metropolis, WalkSAT, genetic algorithms) have been successful as optimization heuristics in a number of application areas. However, very little is known about which problems they succeed at, which instances they do well on, or how to make implementation choices to tailor the method to a given application. To address this, recent theoretical work on population-based methods will be used as a guide in an experimental study of the structure of search spaces. Population- based algorithms (in particular go-with-the-winners local optimization, will be used to sample uniformly from the part of the search space that meets a certain minimal standard of optimality. Such samples will be used to discover combinatorial characteristics of search spaces, and then to predict and improve the performance of optimization heuristics. Social Impact: There will be collaboration between theorists and experimenters. This will help forget ties between the two research communities. Graduate students will learn to balance theoretical skills, programming institution, and concern for applications