Carl Smith - US grants

The goal of the Capital Area Theory Seminar is to stimulate research activity in theoretical computer science. The plan is to have a talk every other week at the University of Maryland, College Park, from fall of l991 to spring of 1994. The requested funds, with matching funds from University of Maryland Institute for Advanced Computer Studies, will be used to invite ten outside speakers per year.

This project will contribute toward the goal of understanding how a computer can be programmed to learn by isolating features of incremental learning algorithms that theoretically enhance their learning potential. The features are chosen to resemble features of human learning. For example, learning devices that use the results of one learning episode to aid in the next learning endeavor will be considered. Another powerful learning technique, for both humans and machines, is the ability to ask questions. Investigations into the relative learning potential of algorithms as the language they use to pose questions will be conducted. Preliminary results indicate that the more powerful the query language used, the greater the learning potential of the device asking questions. Parameters of query language strength such as the number of quantifiers and the number of alternations of quantifiers will be vigorously examined. Current program testing techniques do not and cannot determine correctness precisely. The purpose of these techniques is to reveal errors in a program; however, if no errors are found, no measure is given as to how reliable the program is. Therefore there is no notion of how "close" to correct the tested program is. The goal of this research is to develop a theory whereby it is possible to say precisely how reliable a program is when it passes a given test.

9401842 Khuller Over recent years, the Capital Area Theory Seminar has been particularly active, hosting a talk every other week. Some of the speakers for the seminar are local, and others are visitors to the DC area. The mailing list has 107 entries, some of which are other mailing lists. The goal of the seminar is to stimulate activity in theoretical computer science. In this regard, the seminar has been quite successful. The seminar is often attended by researchers from Univ. of Maryland at Baltimore County, Johns Hopkins Univ. and Georgetown University. ***

Various aspects of machine learning have been under empirical investigation for quite some time. And, more recently, theoretical studies have become popular. This research contributes towards the goal of understanding how a computer can be programmed to learn by isolating features of incremental learning algorithms that theoretically enhance their learning potential. In addition to pursuing lines of research that have already been established with a firm base, investigations of other topics in machine learning are initiated; for example, investigating learning algorithms with a limited ability to remember observed data and other information. Inductive inference is in some sense the dual of program testing. For the inference problem, one must generalize from a finite set of examples to an entire function or concept. The testing problem requires the formulation of a finite test set that distinguishes a given function from all the others in a given class. The program testing problem is more difficult than the program equivalence problem, hence it is highly unsolvable. Consequently, practitioners are doomed to ad hoc techniques. However, it is possible to discuss testing strategies that give an indication of how confident we are that the program being tested is reliable. In another vein, the complexity of certain problems is studied. Some of the earlier work in recursive graph theory has now been applied to the complexity of NP-hard problems. This work is extended to problems in P. The complexity of Boolean functions is also investigated.

Basic research in theoretical computer science is studied. Fundamental questions, about the analysis of algorithms and complexity theory are researched. This seminar series focuses on inviting distinguished outside speakers who are able to give an overview of a research area to first and second year graduate students. It is hoped that the interaction between graduate students, researchers in the DC metropolitan area, and outside visitors will be very productive and useful in stimulating new research and facilitating new collaborations. Students will also get an opportunity to present their work to the faculty. A special effort will be made to involve honors undergraduate students in this endeavor.

The models used in inductive inference have their roots in the models used by the philosophers of science who were discussing the scientific method. The goal there, and in prior work in learning theory, was to come up with an explanation of the phenomenon under consideration. However, scientists rarely work directly for the grand goal of a complete explanation, seeking rather the more modest goal of finding features and facts about the observed data. In like manner, this project pursues several modifications to the traditional models used in inductive inference so as to study the pursuit of more modest scientific goals. This study is particularly relevant to contemporary science as automated data generation techniques produce sufficient volumes of data to overwhelm the analysis abilities of humans. The goal of our work is to illuminate precisely what can and cannot be accomplished by automatic data analysis algorithms.

he models used in inductive inference have their
roots in the models used by the philosophers of
science who were discussing the scientific method.
The goal there, and in prior work in learning theory,
was to come up with an explanation of the phenomenon
under consideration.
However, scientists rarely work directly for the grand goal
of a complete explanation.
The more modest goal of finding features and
facts about the observed data is pursued.
A variety of types of algorithms that could be construed as
discovering their final result will be investigated.
We propose to consider computations that discover rather than compute
their intended result.
A logic of discovery will be developed and investigated.
This study is particularly relevant to contemporary
science as automated data generation techniques
produce sufficient volumes of data to
overwhelm the analysis abilities of humans.
The goal of our work is to illuminate
precisely what can and cannot be accomplished
by automatic data analysis algorithms.
Such algorithms are used in data mining and
text analysis for world wide web search engines.

DESCRIPTION (provided by applicant): The early postpartum period is an extremely demanding time for new mothers. Most women remain psychologically healthy but emotional disorders, such as high anxiety, are common. Anxiety disorders left untreated can have both short and long term detrimental consequences for the mother and infant. The neural substrates leading to stable or improved mood for most new mothers, or elevated anxiety for other mothers, are poorly understood. Laboratory rats are an excellent model to examine postpartum changes in anxiety-related behaviors because they, like most women, experience a decrease in anxiety during the early postpartum period. Our laboratory has recently demonstrated that this change is partially dependent on the tactile stimulation dams receive from their litter. Indeed, preventing lactating females from accessing their pups for as little as four hours increases dams'anxiety to that of diestrus virgins. Recent experiments I conducted suggest that the ventral bed nucleus of the stria terminalis (BSTv) is important for maintaining this attenuated anxiety in dams allowed contact with their pups. The BSTv is neurochemically unique because it contains one of the densest noradrenergic innervations in the brain. The proposed experiments will, thus, examine the anxiety-related effects of norepinephrine within the BSTv of lactating female rats. The first specific aim will determine if noradrenergic agonists or antagonists infused directly into the BSTv alter anxiety-related behaviors in females either allowed access to their pups, or not, before exposure to an anxiogenic test. Second, levels of norepinephrine within the BSTv will be monitored with microdialysis before and after dams are separated from their pups. Finally, the third specific aim will use triple-label immunohistochemistry to determine if modulated GABAergic neurons within the BSTv project to the ventrocaudal periaqueductal gray (cPAGv) after exposure to an anxiogenic stimulus. The cPAGv is a "final common pathway" for anxiety-related behaviors and may need to be inhibited by the BSTv for reduced anxiety-related behaviors. These specific aims will determine: 1) if altering norepinephrine within the BSTv affects anxiety, 2) if norepinephrine levels change as the result of contact with pups, and 3) if the BSTv modulates a larger fear circuitry in the brain. Knowledge gained from these experiments will elucidate the neural mechanisms underlying maternal anxiety in new mothers and enhance the diagnosis and treatment of anxiety disorders in postpartum women.