Our group meets on Thursdays at 11:00 a.m., FL 246.
- Jan 27
- Feb 3
- Dovan Rai
- "Causal Modeling of user data from Monkey's Revenge: a math
tutor with game-like elements"
We have used causal modeling to understand data from a math
learning environment with game-like elements, Monkey's Revenge. We
collected data from 297 middle school students in various categories
such as their attitude, enjoyment, performance and learning. Although
the data are observational, we explored causal modeling approaches and
have obtained some interesting results that not only confirmed our
prior hypotheses about data but also generated interesting hypotheses
that would not have been possible to induce from statistical methods
alone. Based on our comparison of this relatively new approach with
traditional statistical tools like correlation and multiple
regression, we found that causal modeling infers direct, indirect and
spurious associations between variables enabling us to understand
interrelationships within the overall data.
- Feb 10
- Feb 17
- Academic Advising Appointment Day
- Feb 24
- March 3
- Peter Swire
- "A First Look At Dempster-Shafer Theory"
There are times when we must reason under uncertainty:
sensors pick up noise, mistakes get recorded, and experts disagree.
How can you draw conclusions from data that seems to contradict
itself? Dempster-Shafer theory is a collection of mathematical tools
for reasoning over evidence that cannot be fully trusted. This week's
AIRG will be a brief and gentle introduction to it.
- Mar 10
- Mar 17
- Mar 24
- Yue Gong
- "Looking beyond transfer models: finding other sources of power
for student models"
Student modeling plays an important role in educational research. Many
techniques have been developed focusing on accurately estimating
student performances. In this paper, using Performance Factors
Analysis as our framework, we examine what components of the model
enable us to better predict, and consequently understand, student
performance. Using transfer models to predict is very common across
different student modeling techniques, as student proficiencies on
those required skills are believed, to a large degree, to determine
student performance. However, we found that problem difficulty is an
even more important predictor than student knowledge of the required
skills. In addition, we found using student proficiencies across all
skills works better than just using those skills thought relevant by
the transfer model. We tested our proposed models based on two
transfer models with fine and coarse grain sizes; the results suggest
that the improvement is not simply an illusion due to possible
mistakes in associating skills with problems.
- Mar 31
- Jason Wilson (BAE systems)
- "Hybrid Qualitative Simulation of Military Operations"
Our goal is to enable military planners to rapidly critique
alternative battle plans by simulating multiple outcomes of
adversarial plans. We describe a novel simulator, SimPath,
that combines qualitative reasoning, a geographic
information system (GIS), and targeted probabilistic
calculations to envision how adversarial battle plans can
play out. We outline the problem and describe the overall
operation of the simulator. We then explain how qualitative
process theory is extended with actions to model military
tasks, how envisioning is factored to reduce combinatorial
explosions, and how probabilities are computed for
transitions and used to filter possibilities. Empirical results,
including an experiment conducted by an independent
evaluator, are summarized. The results show that it is
possible to identify dozens of possible outcomes on each of
9 combinations of adversarial plans (COAs) in under two
minutes. We close with a discussion of future work.
- Apr 07
- Shubhendu Trivedi, Zachary A. Pardos, Neil T. Heffernan
& Gabor N. Sarkozy.
- "The Utility of Clustering in Prediction Tasks"
In this work we explore the utility of clustering data to aid in
prediction tasks and propose a simple yet very effective bagging
meta-algorithm for prediction. In hindsight the work can be thought of
something like an adaptive mixture of experts (Hinton, Jordon 1991)
that uses clustering to bootstrap. But unlike in other bagging
methods, which select a random subset to bootstrap, this method has a
specific expert (a predictor) for a specific cluster of the data. By
varying the granularity of the clustering we are able to obtain a set
of diverse predictions on the test set that are then combined together
to get a single much stronger prediction. The single most important
aspect of the method is clustering. Methods such as k-means are based
on estimating explicit models of the data such as a mixture of
spherical gaussians. They perform badly when these assumptions are not
met, which is usually the case. When it comes to more complex and
unknown shaped data distributions, spectral clustering methods are
known to return much stronger results. We also compare results on a
few benchmark datasets using both k-means and spectral clustering.
- Apr 14
- Apr 22
- Undergraduate Project Presentation Day
- Apr 28
- May 2
- Spring semester graduate classes end