Tetewsky, S., Shultz, T., & Buckingham, D. (1993). Reducing retroactive interference in connectionist models of recognition memory. Canadian Society for Brain, Behaviour, and Cognitive Science, Third Annual Meeting, 93.



Previous research has shown that the Back-Propogation learning algorithm (BP) cannot model human performance in a serial recognition memory task because it produces excessive amounts of retroactive interference (i.e. "catastrophic" interference.)   Using an expanded version of the paradigm introduced by Ratcliff (1990, Experiment 1), three sets of simulations were run to evaluate the generalizability of this conclusion.  The first simulation showed that when the amount of interference was measured implicitly, in terms of the number of training epochs required to relearn a given pattern set, BP retained a minimum of 80% of the information it had originally learned.  Two additional sets of simulations were then carried out to determine if the Cascade-Correlation learning algorithm (CC), which constructs its own network topology by adding hidden units as needed, could improve on this finding.  The first CC model saved as much as 90% of the information it originally learned, but it also had certain limitations due to a design feature of its architecture.  The second CC model overcame these limitations while also saving over 95% of the information it originally learned.  In addition, it made several psychological predictions suggesting that a generative learning algorithm like CC may be better able to account for human performance than BP. 


Copyright notice

Abstracts, papers, chapters, and other documents are posted on this site as an efficient way to distribute reprints. The respective authors and publishers of these works retain all of the copyrights to this material. Anyone copying, downloading, bookmarking, or printing any of these materials agrees to comply with all of the copyright terms. Other than having an electronic or printed copy for fair personal use, none of these works may be reposted, reprinted, or redistributed without the explicit permission of the relevant copyright holders.