Paradoxical replay can protect contextual task representations from destructive interference when experience is...

Area of Science:

• Neuroscience
• Computational Neuroscience
• Machine Learning

Background:

• Experience replay is crucial for efficient learning, but the selection criteria for replayed experiences are not fully understood.
• A puzzling observation is that rats replay less-experienced trajectories when experience is unbalanced.
• Current theories struggle to explain this paradoxical replay behavior.

Approach:

• Simulated a feedforward neural network with 'rich' (task-specific) and 'lazy' (task-agnostic) learning regimes.
• Investigated how unbalanced experience affects these representations and whether paradoxical replay mitigates degradation.
• Examined the link between paradoxical replay and hippocampal representations in rats.

Key Points:

• Rich representations degraded under unbalanced experience, while lazy representations did not.
• Paradoxical replay reversed the degradation of rich representations.
• A strong association exists between the richness of learned representations and paradoxical replay in rats.

Conclusions:

• Paradoxical replay protects rich neural representations from the detrimental effects of unbalanced experience.
• This study demonstrates a novel interaction between the nature of task representations and the function of replay in both artificial and biological systems.