Related Experiment Videos
Effect of memory decay on predictions from changing categories
1Department of Psychology, Carnegie Mellon University, Pittsburgh, Pennsylvania 15213, USA.
Summary
Real-world categories change over time. This study shows people adapt to category shifts, with memory decay influencing their learning, supporting exemplar-based categorization models.
Area of Science:
- Cognitive Psychology
- Computational Neuroscience
- Machine Learning
Background:
- Traditional categorization research often assumes static categories.
- Real-world categories, however, are dynamic and evolve over time.
- Understanding adaptation to changing categories is crucial for cognitive models.
Purpose of the Study:
- To investigate how individuals adapt to gradually changing categories.
- To model the effects of past observations and memory decay on categorization performance.
- To determine the nature of the memory decay function in dynamic categorization.
Main Methods:
- Four experiments were conducted to observe participant behavior during category change.
- Computational models, including rational categorization and exemplar-based models, were employed.
- Memory decay was incorporated into existing categorization algorithms (Anderson, 1990, 1991; Nosofsky, 1986).
Main Results:
- Participants successfully adjusted to category changes.
- A lingering, cumulative effect of past observations was evident.
- Models incorporating memory decay accurately predicted performance.
- The memory decay function approximated a power law, stronger for items than time.
Conclusions:
- Human categorization dynamically adapts to evolving category definitions.
- Memory decay plays a significant role, with power-law characteristics.
- Findings support the use of exemplar memories in categorization processes.
- Computational models provide valuable insights into dynamic category learning.