The effects of the post-delay epochs on working memory error reduction.

Area of Science:

• Cognitive Neuroscience
• Computational Neuroscience
• Machine Learning in Neuroscience

Background:

• Accurate information retrieval is vital for working memory, particularly during post-delay epochs involving cues and responses.
• The computational and neural mechanisms supporting robust memory during these critical post-delay periods are not well understood.

Purpose of the Study:

• To investigate the computational mechanisms underlying robust working memory retrieval during post-delay epochs.
• To explore how neural networks adapt to environmental statistics to improve memory performance.

Main Methods:

• Training recurrent neural networks (RNNs) on a color delayed-response task with varying stimulus frequencies.
• Analyzing neural activity patterns and decoding strategies within the RNNs during post-delay epochs.
• Investigating the role of neural dynamics and readout processes in memory error reduction.

Main Results:

• Trained RNNs demonstrated reduced memory errors for frequently presented 'common' colors.
• This error reduction was achieved by decoding a wider range of neural states into common colors during post-delay epochs.
• The decoding process involved convergent neural dynamics and a biased, non-dynamic readout mechanism.

Conclusions:

• Post-delay epochs are crucial for working memory, enabling adaptation to environmental statistics.
• Neural systems employ multiple mechanisms, including altered decoding strategies and readout biases, to enhance memory robustness.
• RNN models provide insights into the computational principles governing working memory and neural adaptation.