AIRG Topics - Fall 2009

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 10

AIRG Organizational Meeting (Coordinator: Joe Beck)

Sept 17

Joseph E. Beck

"Using machine learning to better understand human learning"

This talk focuses on the capabilities granted us by large datasets for better understanding human learning. Typically modelers attempt to model average performance of the population, and neglect what an individual trial looks like. This talk will explore the potential benefits of examining learning at a finger grain-size. Two hoped for outcomes are a better fit to actual human performance data (and thus, a better model of human learning), and an ability to determine which events occur just before dramatic shifts in student knowledge. This talk is applicable to those not interested in human learning as a topic area, as many researchers (AI and otherwise) have to choose between modeling data at the individual level or at the aggregate.

Sept 24

Elijah Forbes-Summers

"Possible Research Directions"

I am going to talk about some possible directions for my thesis, all involving the Assistment system. There are three possibilities. The first involves unsupervised reinforcement learning over logs from the system in order to discover novel relationships; the second is an extension of the current effort on the bar and teacher dashboard prototype that allows teachers to monitor students progress (and engagement); the third possibility focuses on the role of students in the learning process, specifically looking at giving students certain freedoms within the system and exposing them to the consequences of their actions.

Oct 1

Paul Gibler POSTPONED

AIRG members are encouraged to attend the joint CS Colloquium and IMGD Speaker Series talk by Dr Eva Hudlicka on affective gaming.

Oct 8

(no meeting)

Oct 15

Dovan Rai

While the motivational benefits of computer games are appealing to education designers, when it comes to learning gains they are still inferior to intelligent tutors . Hence incorporating the features of games that are motivational and that also help learning (at least do not hurt) into tutors seems to be the next good option. In this talk, I want to explore what are the game-like properties in general and in context of tutors. I will give two examples where we have implemented these properties: one in existing tutor, assistment and another flash based math tutor.

Oct 22

Ugrad break - No meeting

Oct 29

Yue Gong and Elijah Forbes-Summers talk POSTPONED

Nov 5

Paul Gibler's talk POSTPONED and replaced by...

Dave Brown

Group Discussion: "If Computational Artistic Creativity is Possible How Would We Start?"

Computational Creativity researchers are working on producing art, writing and designs. There are a lot of issues involved with producing artifacts that people would judge to be creative. Can computers judge whether something is creative? Can they judge an artistic artifact in particular? How would we start research to produce creative art? What goals might such an effort have? What are the main barriers? This will be a guided but relatively free-form discussion.

Nov 12

Adam Goldstein

"Transitioning to the Transition Variable; Modeling Contextual T in Bayesian Knowledge Tracing"

Since Corbett and Anderson's work in 1995, significant progress has been made on tracing student learning with Bayesian Knowledge Tracing in Intelligent Tutoring Systems. Beck recognized the issue now known as the Identifiability Problem, where identical data can be modeled to different parameters, and Baker, Corbett, and Aleven contextualized the guess and slip variables using Machine Learning. Using a similar process, it is perhaps possible to also contextualize the transition variable to both improve BKT performance and to help make reportable knowledge to users in ITSs available and accurate.

Nov 19

Yue Gong

"The impact of gaming on learning" (part 1)

One of the common expectations of educational researchers, ITS designers and school teachers is that students learn efficiently from every practice opportunity. However, when students are using intelligent tutoring system, there are many different types of non-learning behaviors, such as "Gaming the system", which is considered the one that can strongly reduce learning [Baker 2004.]. In this study, we modified Knowledge Tracing model [Corbett and Anderson 1995] to make inference about the students' initial knowledge and learning given the information of students' gaming state. Moreover, we offered an automatic generic mechanism to evaluate any gaming detector's precision rate by only using data on hand. Our results show learning rate is much lower when students are gaming and less initial knowledge causes more gaming behaviors.

Nov 26

Thanksgiving Break - No Meeting

Dec 3

Yue Gong

"The impact of gaming on learning" (part 2)

One of the common expectations of educational researchers, ITS designers and school teachers is that students learn efficiently from every practice opportunity. However, when students are using intelligent tutoring system, there are many different types of non-learning behaviors, such as "Gaming the system", which is considered the one that can strongly reduce learning [Baker 2004.]. In this study, we modified Knowledge Tracing model [Corbett and Anderson 1995] to make inference about the students' initial knowledge and learning given the information of students' gaming state. Moreover, we offered an automatic generic mechanism to evaluate any gaming detector's precision rate by only using data on hand. Our results show learning rate is much lower when students are gaming and less initial knowledge causes more gaming behaviors.

Dec 10

Amro Khasawneh

Cancelled

Dec 17

Steve Giguere: "Using dynamic bayesian networks to model gaming"

Sugandha Goyal: "Model for designing examinations"

Yutao Wang: "Student Performance Prediction using Bayesian Networks"