AIRG Topics - Fall 2002

Our group meets on Thursdays at 11:00 a.m., FL 246.

Dates and topics for this semester are as follows:

Sept 5

AIRG/DKBRG Organizational Meeting (Coordinators: DCB & EAR)

Sept 12

Dan Dougherty

"Unification and Matching"

Unification problems are problems of the form: "given two patterns, do they have a common instance?" A matching problem is given by a pattern and a term,where the question is, "is this item an instance of this pattern?"

Different precise notions of unification and matching arise once we specify what we mean by "term", "pattern", and "instance". Then, these different varieties of unification and matching find applications in database querying, automated deduction, logic programming, program transformation....

In this talk I'll give the basic definitions and an overview of the applications, then describe some of the particular questions I have been interested in recently.

Sept 19

Cancelled

Sept 26

Neil Heffernan

"Data-Mining to Improve Intelligent Tutoring Systems: Using Search and machine learning"

Intelligent tutoring systems generate a large amount of data as students used them. I will discuss a data-mining problem, which is just beginning to be explored. Current systems have models that predict student difficulty on individual classes of problems. The research question is, "How do we mine this data, so we can improve these tutoring systems?" More specifically, "Does the data on student learning conform to the models already in these tutors, and if not, can we build better models?" Our current solution combines AI search techniques with statistical analysis (currently using logistical regression). In particular we search for the best fitting and most parsimonious model to explain the data. A key idea that we leverage is the Power Law of Learning, which states that error rates & latency rates can be modeled as conforming to a power function. I will share some initial results and hope to interest you in this "new" type of data-mining problem. This topic could make a good master's thesis topic for a student with experience with statistics and an interest in machine learning.

Oct 3

End of A term

Oct 10

Carolina Ruiz

"Discovering Correlations in Sequential Data"

Oct 17

Term break

Oct 24

Tracy A. Hammond, MIT AI Lab.

"Sketch Recognition in Software Design and Other Domains"

Sketching is a natural and integral part of software design. Software developers use sketching to aid in the brainstorming of ideas, visualizing of program organization, and understanding of requirements. We want to allow software design meetings to continue as they are, with software designers discussing the design and drawing free-hand sketches of these designs on a whiteboard. We use sketch recognition to recognize and interpret these designs without interfering with the current design process. Because the diagrams are interpreted, we provide natural editing capabilities to the designers, allowing the users to edit their original strokes in an intuitive way. The interpreted diagrams are used to automatically generate stub code using a software engineering tool. Software design meetings are videotaped to capture visual and spoken design information unobtrusively. When drawn items are interpreted, we use these understood sketch events to index the videotape of the software design meeting.
     Software design is not the only domain that can benefit from sketch recognition. We wish to allow users to add sketch recognition to their system without implementing the sketch recognition code details themselves. For this we propose a domain description language used to describe domain-specific information to a domain-independent sketch recognition system. Although the language is primarily based on shape, the domain description can include any type of information that would be helpful to the recognition process, such as stroke order or direction. The language consists of predefined shapes, constraints, and editing behaviors, as well as a syntax for specifying a domain description.

Oct 31

Aparna Varde

Presentation and discussion of:

Agma Traina, Caetano Traina, Spiros Papadimitrou & Christos Faloustaos

"Tri-Plots: Scalable Tools for Multidimensional Data Mining"

In Proceedings of: SIGKDD 2001.

We focus on the problem of finding patterns across two large multidimensional datasets. For example, given feature vectors of healthy and of non-healthy patients, we want to answer the following questions: Are the two clouds of points separable? What is the smallest/largest pair-wise distance across the two datasets? Which of the two clouds does a new point (feature vector) come from? We propose a new tool, the tri-plot, and its generalization, teh pq-plot, which help us answer the above questions. We provide a set of rules on how to interpret a tri-plot, and we apply these rules on synthetic and real datasets. We also show how to use our tool for classification, when traditional methods (nearest neighbor, classification trees) may fail.

Nov 7 {location changed to FL 320}

Ph.D. students from U. Sydney

Udo Kannengiesser

"Situated Product Modelling"

Designing is a process whose most distinguishing feature is that it changes the world in which it takes place. Appropriate models of designing therefore assume an open, dynamic world in which the design agent's knowledge is grounded, as opposed to a pre-defined, static world where all knowledge is encoded. Situatedness is the notion that describes this dynamic nature of designing and forms the basis for developing design agents.
     This research concentrates on the communication between situated design agents. It argues that existing approaches to communication in multi-agent systems (MAS's) are not appropriate in the design domain, as they are based on static ontologies that are not in accord with the paradigm of situatedness. We aim to develop a situated model of communication, which builds on the notion of common ground, a central concept in cognitive studies of human language use. Common ground can be seen as the shared background of the agents involved in communication, however it the forms part of the grounded knowledge of the individual agent. As a result, common ground is non-static and can be used by the agent to adapt the production and comprehension of messages to the agent(s) it communicates with. We present our current work on the integration of common ground into an architecture of a situated design agent.

Pak-San Liew

"Constructive Memory in Design"

The inclusion of the notion of situatedness into the construction of computer aided design system facilitates the creation of computational systems that have the ability to operate in an environment for which they have not been directly programmed. Situatedness in designing entails the explicit consideration of the effects of the current external environment, the internal state of a design agent; and the interaction between the environment and the agent, on the behaviour of a design agent in performing a design task. It employs the notions of a design situation and of constructive memory.
     This talk describes the theory of constructive memory in design. An implementation model of the constructive memory system is also illustrated together with some engineering applications.

Nov 14 {location changed to FL 320}

Ph.D. students from U. Sydney

Julie R Jupp

"Measuring the Information Content of Architectural Plans"

Humans have long used ordering principles that allow various domains of individual phenomenon and systems to be structured. Reasoning about things individually is difficult without first classifying or measuring them according to distinguishable properties. In many domains, phenomenon and systems are often presented in terms of sequences or strings of characters allowing access to properties and enabling the measurement of certain observable, i.e., sequences of data.
     This talk describes an approach to the measurement of the information content of two-dimensional design drawings. A general method for extracting information from an encoded string of symbols as a canonical representation of architectural plans is presented. Based on the computation of the remoteness of two bodies of knowledge the key point of this model is the qualitative encoding of shape features in order to accurately calculate tThe information content of each drawing is , determined by measuring its entropy. The information content of design drawings provides the basis for measuring shape complexity, for comparing and categorising 2D plans automatically and lays the foundation for the development of a digital representation of style.

Ricardo Sosa

"Implementation of the a Social Model of Creativity in Design"

There is growing advocacy for the adoption of computational methods as a substitute for, or complement to, traditional research methods, particularly for examining social phenomena derivative of organised complexity. This paper examines some of the reasons for this advocacy and the specific advantages of the method for studying such phenomena. It considers also the limitations and problems that need to be addressed if the method is to gain wider acceptance. In joining in the advocacy of these techniques, a framework is proposed which can assist with the incorporation of computational techniques in a broader methodological mix. Such a mix has the potential to harness the strengths of the method while offsetting some of its weaknesses.

Nov 21

Zhuo Chen

"Exploring a Two Layered A-Design System"

MS Thesis presentation

The system described configures components to satisfy constraints, producing new components if performance is inadequate. It is an agent-based system with behavior similar to a Genetic Algorithm. Results are presented and performance evaluated.

Nov 28

Thanksgiving Break

Dec 5

Geraldine Rosario

"Providing order and distance to nominal variables via Focused Correspondence Analysis"

Ph.D. Qualifying Exam Directed Research Report

Dec 12

No meeting