Structured experience shapes strategy learning and neural dynamics in the medial entorhinal cortex

Area of Science:

• Neuroscience
• Computational Neuroscience
• Animal Behavior

Background:

• Cognitive flexibility and generalization are crucial for animals to solve novel tasks.
• The structure of prior experience, particularly early training, influences learning and adaptive behavior.
• Neural mechanisms underlying how early training shapes learning remain largely unknown.

Purpose of the Study:

• To investigate how the structure of early training influences learning, generalization, and neural mechanisms.
• To compare learning and generalization in recurrent neural networks (RNNs) and mice under different training curricula.
• To identify neural dynamics that support flexible and generalizable knowledge.

Main Methods:

• Trained RNNs on variants of an odor-timing task with different curriculum structures.
• Conducted behavioral and electrophysiological recordings in mice trained on the same task with staged sequences.
• Applied dynamical systems approaches to analyze neural activity patterns and network dynamics.

Main Results:

• RNNs and mice with poorly structured early training exhibited rigid strategies and frequent errors.
• Animals and networks with balanced early training demonstrated enhanced generalization and adaptive behavior.
• Structured curricula led to distinct dynamical motifs in networks, facilitating correct abstractions and generalization.

Conclusions:

• The structure, not just the presence, of prior experience is critical for developing flexible and generalizable knowledge.
• Early training curriculum design significantly impacts the emergence of adaptive behavior and neural representations.
• Findings provide insights into the neural basis of learning and generalization in both biological and artificial systems.