Related Experiment Videos
Connectionist models of recognition memory: constraints imposed by learning and forgetting functions
1Psychology Department, Northwestern University, Evanston, Illinois 60208.
Psychological Review
|April 1, 1990
Summary
Connectionist models struggle with sequential memory tasks, rapidly forgetting old information and showing poor discrimination between learned and new items. Current models and variants fail to resolve these fundamental limitations.
Area of Science:
- Cognitive Science
- Computational Neuroscience
- Artificial Intelligence
Background:
- Multilayer connectionist models utilizing backpropagation are common for memory research.
- These models are applied to recognition memory tasks with sequential item presentation.
Purpose of the Study:
- To evaluate multilayer connectionist models of memory based on the encoder model.
- To identify and address limitations in these models when applied to sequential learning.
Main Methods:
- Evaluation of standard encoder-based multilayer connectionist models.
- Examination of network manipulations and model variants (prelearned memory, context model).
- Application to sequential recognition memory paradigms.
Main Results:
- Sequential learning causes rapid forgetting of well-learned information.
- Discrimination accuracy between studied and new items degrades or behaves nonmonotonically.
- Tested model modifications and variants did not resolve these core issues.
Conclusions:
- Significant limitations exist for connectionist models in human memory simulation, especially with sequential learning.
- Current models are inadequate for tasks where all information is not available during the learning phase.