AIRG Topics - Fall 2006

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

Dates and topics for this semester are as follows:

Sept 7

AIRG Organizational Meeting (Coordinator: DCB)

Sept 14

Winning The DARPA Grand Challenge -- Video

http://video.google.com/videoplay?docid=8594517128412883394

Sept 21

Sept 28

Kismet Videos

http://www.iwaswondering.org/cynthia_video.html

http://www.ai.mit.edu/projects/sociable/videos.html

Oct 5

Oct 12

Oct 19

No meeting

Oct 26

Dave Brown

Assumptions in Design

When designing software, buildings, gardens, clothes, ...
- What are assumptions for?
- Why do we make them?
- Are they always bad?
- Do we know we're making them?
- How can they be automatically detected or inferred?

Nov 2

IMGD lecture series: Kevin Burns
Salisbury Labs 123

Imagine a community with thousands of people sitting at machines playing games for hours. What makes it fun? Is it virtual reality? Is it engaging narrative? Is it multiplayer interaction? Actually, it's none of the above. The community is Foxwoods and the machines are slots. I'll bet that slots are the most popular and profitable machine game of all time - more than any modern computer game. I also think that research and development in digital media has not done much to advance a scientific understanding of fun in any game. So I cut to the chase, dissecting the aesthetic experience using mathematical analyses and psychological experiments. I look at gambling, music and artwork. I show how formal notions of Bayesian probability and Shannon entropy can explain and predict feelings of pleasure. I have some demos to make the math fun. I guarantee you have never seen fun like this before.

Nov 9

Stuart Floyd

"Applying Data Mining Techniques to Pancreatic Cancer"
This presentation covers initial research into using machine learning algorithms to predict survival of pancreatic cancer patients using a sample of sixty patients treated at the University of Massachusetts Medical Center. Results from tests using Artificial Neural Networks and Bayesian approaches will be presented and compared with results from logistic regression.

Nov 16

Yuan "Ryan" Gao

"Association Rule Mining Algorithm with Weighted Attributes/Items (ARMWAI)"

Traditional association rule mining methods treat all the data in a dataset equally, without considering the fact that attributes/items may vary in importance. For example, in the market basket analysis domain, certain rules like {TV, DVDplayer => DVD} may be more interesting than {HairBrush, Conditioner => Shampoo} since the items in the former rule have higher retail values. Hence, we want to pay more attention to those highly weighted attributes/items, as well as the association rules containing them. Our proposed association rule mining with weighted attributes/items approach is two-fold. First we assign weights to the attributes/items in the dataset. Different notions of importance may be used to obtain these weights. Then we modify the traditional Apriori association rule mining algorithm to mine association rules from the weighted dataset. We describe an application of this approach to bioinformatics, the domain which originally motivated this research.

Nov 23

Thanksgiving Break
Nov 30

Darren Torpey

"Reducing the cost of creating ITS pseudo-tutors via a web-based interface"

MS Thesis Presentation

Intelligent tutors have been shown to be effective ways of teaching children grade-school math. Unfortunately, it has been estimated that it takes between 200 and 300 hours to create a single hour of intelligent tutoring content for a student. WPI and CMU have been funded by the Office of Naval Research to explore ways to reduce the cost associated with creating cognitive model-based tutors used in Intelligent Tutoring Systems (ITSs). Traditional ITSs have been built by programmers who need PhD-level experience in AI rule-based programming as well as backgrounds in cognitive psychology. WPI is taking a different approach, hoping to lower the skills needed to the point that normal classroom teachers can author their own ITS content. We have collected data on the amount of time it takes people to author tutors in ASSISTments. Turner reported, based on the creation of 25 tutors, that ASSISTments reduced the cost by a factor of at least 4, with a ratio of 45:1. Since the ASSISTments System has over 300 tutors being used in schools, his estimates may not accurately reflect the time it takes to make a real tutor in ASSISTments. We will be collecting data on at least 125 ASSISTments to see if his predicted ratio of 45:1 is accurate within a real content-production process for grade-school algebra tutors.

Dec 7

** Presentation Postponed **

** New date/time: Tuesday 12th Dec., 4-5pm **

** Location: FL 246, Beckett Conf. Room **

Kai Rasmussen

"Developing a Cognitive Rule-based Tutor for the ASSISTment system"

MS Thesis Presentation

The ASSISTment system is a web-based tutor that is currently being used as an eighth and tenth-grade mathematics in both Massachusetts and Pennsylvania. This system represents its tutors as state-based "pseudo-tutors" which mimic a more complex cognitive tutor based on a set of production rules. It has been shown that building pseudo-tutors significantly decreases the time spent authoring content. This is an advantage for authoring systems such as the ASSITment builder, though it sacrifices greater expressive power and flexibility. A cognitive tutor models a student's behavior with general logical rules. Through "model-tracing" of a cognitive tutor's rule space, a system can find the reasons behind a student action and give better tutoring. Also, these cognitive rules are general and can be used for many different tutors. It is the goal of this thesis to provide the architecture for using cognitive rule-based tutors in the ASSITment system. A final requirement is that running these computationally intensive model-tracing tutor do not slow down students using the pseudo-tutors, which represents the majority of ASSISTment usage. This can be achieved with remote computation, realized with SOAP web services.

Dec 14

No Meeting