 |
|
|
|
|
|
|
|
|
|
|
|
|
 |
|
|||||||||||
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
|
|
 |
 |
|
||||||||||
 |
|
|||||||||||
 |
|
|||||||||||
 |
 |
|
||||||||||
| Seminar Schedule (Seminar Organizer: Prof. Ziv Bar-Joseph) | |
Machine Learning Lunchtime Chats
|
Fall 2007 Seminar
Date: December 19, 2007 Time: 4:00 PM - 5:00 PM Location: 4623 Wean Hall Speaker: Hao Cen Ph.D. Candidate Title: Generalized Learning Factors Analysis: Improving Cognitive Models Abstract: Cognitive Tutors are award-winning computer-based math curriculums that grow out of the extensive research in human learning and artificial intelligence at Carnegie Mellon. Evidence indicates that students using these curriculums perform significantly better on several standardized tests and problem solving tests. By 2007, more than 500,000 middle school students began using Cognitive Tutors across the United States. The full potential of Cognitive Tutor has not yet been reached, though. The issues mainly concern the cognitive models used, which describe a set of skills that represent how students solve domain problems. However, due to the large quantity and the latent nature of skills encoded in a cognitive model, and "expert blind spots" from cognitive model authors, the existing cognitive models are usually an incomplete representation of student knowledge, resulting in both less accurate assessment of student knowledge and lower student learning efficiency than desired. Improving the existing cognitive models, given the rate at which Cognitive Tutors are used across the U.S., has an immediate and significant impact on student learning, and has a long-term impact on transforming math curriculum design. To address this challenge, we developed a machine learning framework called Learning Factors Analysis (LFA). By combining human learning theory, machine learning technology, and psychometrics, this framework can evaluate existing cognitive models against student learning data and semi-automatically search for better models. It paves the way for researchers and practitioners to fine-tune cognitive models on solid student data. In the thesis proposal, I generalize LFA by 1) developing Conjunctive Factor Model to predict student performance more accurately than the early LFA model on items requiring multiple skills; 2) using the new model and the combinatorial search to find a convincingly better and interpretable cognitive model in a real assessment environment; 3) using Exponential PCA to identify latent factors and compare their predictive performance with expert labeled factors ; 4) Implementing a faster parameter estimation and model searching algorithm for its coming uses with large data sets. Thesis Committee: Ken Koedinger (Chair), Geoff Gordon, Brian Junker, Noel Walkington (Math Dept.) http://www.learnlab.org/uploads/mypslc/publications/thesis%20proposal%201.1. pdf