Research Projects
-
Question/Answering Systems
The goal is to develop methods for answering complex queries about
a number of domains of interest using answer set programming.
Joint work with Chitta Baral and Michael Gelfond. (Funded
by ARDA
-
Information/Knowledge Fusion
The goal of this
work is to develop methods for utilizing the vast amounts of data and
information available in different formats in databases and on web pages. My
current work is directed towards the development of a high-level language
(based on Golog) for the specification of information-gathering plans. (Funded
by the Knowledge Fusion Center of the Army Research Laboratory
- Cognitive Robotics:
Intelligent agents acting in the world need to be able to accomplish their
goals without having complete knowledge of their environment. Their plans must
include steps to acquire the needed knowledge. For example, in order to know
whether or not there is an object of a particular shape within its grasp, it
is likely that a robot will have to perform one or more sensing operations.
Furthermore, the ability of an agent to perform an action may depend crucially
on the agent's knowledge. Consider the action of opening a safe by dialing the
combination. A prerequisite of being able to perform this action is knowing
the combination of the safe. Agents also need to reason about the knowledge
and abilities of other agents. For example, the correct use of a referring
expression requires that the robot know that the hearer is able to identify
the referent. The focus of my work has been the logical foundations for
representing and reasoning about actions and their effects. The idea is to
specify with mathematical precision our common-sense knowledge of actions and
how occurrences of various actions affect the state of the world. More
specifically, I've looked at the use of the situation calculus (a particular
logical formalism) in modeling actions that affect the epistemic states of
agents. Part of this work has involved the development of automated methods
for drawing conclusions from such an axiomatization of actions and their
effects. Another aspect of the research is the development of a high-level
programming language (GoLog) for the specification of agents (e.g. robots,
softbots) based on such a situation calculus axiomatization.
- Automated Deduction for Modal
Logics A general framework for the construction of deductive systems that
use first-order modal logic has been developed. The method works by first
translating sentences in modal logic into a constraint logic in which the
constraints represent the accessibility relation in the possible world
semantics for these logics. The framework provides a method for converting
ordinary first-order inference rules into inference rules for constraint logic
by providing a mechanism to incorporate special purpose reasoners to handle
the constraints. A number of existing modal deduction methods can be
reconstructed as instances of this general framework. The result is a new
understanding of how existing methods work and relate to each other.
Additionally, the relationship between sets of inference rules for classical
logic and inference rules for modal logic is made clear. Furthermore, a new
understanding of the relationship between the results of using the functional
translation and the relational translation is produced. In particular, the
E-unification algorithms utilized for reasoning with the result of the
functional translation are seen to be one particular method of solving
constraints. The advantages of the general approach are simple proofs of
correctness for various instances of the framework, applicability to a wide
variety of logics and proof methods, and ease in incorporating additional
features. (Joint work with Alan Frisch)
- The logic of Events An
event space is a set of instantaneous events that vary both in time and
specificity. The concept of an event space provides a foundation for a logical
- i.e., modular and open -- approach to causal reasoning. The goal of this
research is to investigate various axioms for event spaces and methods for
representing the various relations between events including probabilistic
relations. Another aspect of the work is the integration of probability and
logic with the goal of automating such reasoning. (Joint work with Peter
Gillett and Glenn Shafer, funded by NSF)
- Web Mining The goal of this
work is to develop methods for utilizing the vast amounts of data and
information available on the web. Completed work has included a system for
mining information from personal home pages on the Web (joint with James
Geller and funded by the New Jersey Commission on Science and Technology).