Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

One system for learning and remembering episodes and rules

Joshua T. S. Hewson, Sabina J. Sloman, Marina Dubova

TL;DR

The paper addresses whether learning episodic memory and rule generalization require separate systems or can be unified within a single, capacity-robust learner. It demonstrates that by using an associative learning task and varying network capacity, an excess-capacity system can both learn new episodes and rules and retain prior knowledge. This challenges traditional complementary learning systems theories and highlights capacity as a key driver of memory-generalization dynamics. The findings have implications for continual and transfer learning and for designing cognitive architectures that integrate memory and generalization capabilities.

Abstract

Humans can learn individual episodes and generalizable rules and also successfully retain both kinds of acquired knowledge over time. In the cognitive science literature, (1) learning individual episodes and rules and (2) learning and remembering are often both conceptualized as competing processes that necessitate separate, complementary learning systems. Inspired by recent research in statistical learning, we challenge these trade-offs, hypothesizing that they arise from capacity limitations rather than from the inherent incompatibility of the underlying cognitive processes. Using an associative learning task, we show that one system with excess representational capacity can learn and remember both episodes and rules.

One system for learning and remembering episodes and rules

TL;DR

The paper addresses whether learning episodic memory and rule generalization require separate systems or can be unified within a single, capacity-robust learner. It demonstrates that by using an associative learning task and varying network capacity, an excess-capacity system can both learn new episodes and rules and retain prior knowledge. This challenges traditional complementary learning systems theories and highlights capacity as a key driver of memory-generalization dynamics. The findings have implications for continual and transfer learning and for designing cognitive architectures that integrate memory and generalization capabilities.

Abstract

Humans can learn individual episodes and generalizable rules and also successfully retain both kinds of acquired knowledge over time. In the cognitive science literature, (1) learning individual episodes and rules and (2) learning and remembering are often both conceptualized as competing processes that necessitate separate, complementary learning systems. Inspired by recent research in statistical learning, we challenge these trade-offs, hypothesizing that they arise from capacity limitations rather than from the inherent incompatibility of the underlying cognitive processes. Using an associative learning task, we show that one system with excess representational capacity can learn and remember both episodes and rules.
Paper Structure (10 sections, 2 figures)

This paper contains 10 sections, 2 figures.

Table of Contents

  1. Abstract
  2. Introduction
  3. Methods
  4. Catastrophic forgetting.
  5. Model.
  6. Task.
  7. Capacity.
  8. Training.
  9. Results
  10. Conclusion

Figures (2)

  • Figure 1: Temporal plots for mean classification accuracy over training (the noise level is fixed at 25%). Left: The episode (top) or rule (bottom) for $A_{train}-B$ is learned. Right: The episode $A_{train}-C$ is learned (green lines) while the episode (top) or rule (bottom) for $A_{train}-B$ is being forgotten.
  • Figure 2: Final averaged mean results after training on Block 1 and 2 respectively, with varying levels of noise. Left of the dashed line: constrained capacity; Dashed line: sufficient capacity, Right of the dashed line: excess capacity. Error bars show standard errors.