Biologically plausible gated recurrent neural networks for working memory and learning-to-learn
2024

RECOLLECT: A Biologically Plausible Memory Network for Learning-to-Learn

Sample size: 20 publication 10 minutes Evidence: moderate

Author Information

Author(s): van den Berg Alexandra R., Roelfsema Pieter R., Bohte Sander M.

Primary Institution: Centrum Wiskunde & Informatica, Amsterdam, The Netherlands

Hypothesis

Can a biologically plausible gated memory network effectively learn-to-learn and manage memory retention and forgetting?

Conclusion

RECOLLECT successfully learns to represent task-relevant information and can adaptively forget it, demonstrating effective learning-to-learn capabilities.

Supporting Evidence

  • RECOLLECT can flexibly retain or forget information using a single memory gate.
  • It successfully learns to represent task-relevant information over long memory delays.
  • Networks trained with RECOLLECT exhibited learning-to-learn capabilities on a reversal bandit task.
  • Performance improved significantly with the inclusion of an end-of-episode signal.

Takeaway

The RECOLLECT model is like a smart brain that can remember important things and forget unimportant ones, helping it learn better over time.

Methodology

The study involved training the RECOLLECT model on pro-/anti-saccade and reversal bandit tasks to evaluate its memory and learning capabilities.

Limitations

Learning was slower compared to previous models, and the architecture was tested with a single layer of memory units.

Statistical Information

P-Value

0.003

Statistical Significance

p<0.05

Digital Object Identifier (DOI)

10.1371/journal.pone.0316453

Want to read the original?

Access the complete publication on the publisher's website

View Original Publication