On-Line Assessment of Expertise
Abstract
The main objective in this report period was to deal with the combinatorial explosion in the size of the belief networks needed to analyze regular problem solving behavior. As mentioned in our proceeding quarterly progress reports, Olae's belief networks can get quite large and intractable with non-toy knowledge bases. We have three approaches for handling this problem: (1) We improved our algorithm that builds a belief net incrementally so that it is only as large as it needs to be for the particular actions we observe the human to make. (2) We tested a well-known Monte Carlo technique (stochastic simulation, Pearl, 1986) for estimating probabilities on the nodes in very large belief networks. (3) We have developed and tested a technique for simplifying complex Bayesian net structures.
Document Details
- Document Type
- Technical Report
- Publication Date
- Jan 07, 1994
- Accession Number
- ADA274746
Entities
People
- Joel Martin
- Kurt VanLehn
Organizations
- University of Pittsburgh