Attractor dynamics of working memory explain a concurrent evolution of stimulus-specific and decision-consistent...

Area of Science:

• Cognitive Neuroscience
• Computational Neuroscience

Background:

• Sensory information is often transient, necessitating its maintenance in working memory (WM) for cognitive tasks.
• Human perception exhibits systematic biases, including stimulus-specific and decision-consistent biases, in tasks relying on working memory.

Purpose of the Study:

• To elucidate the neural mechanisms underlying stimulus-specific and decision-consistent biases in working memory.
• To propose a unified account for the co-evolution of these biases through dynamic memory processes.

Main Methods:

• Investigated orientation estimation tasks using behavioral experiments and neuroimaging (fMRI).
• Developed and analyzed task-optimized recurrent neural networks (RNNs) to model neural dynamics.
• Employed computational modeling to link drift-diffusion dynamics to observed biases.

Main Results:

• Identified drift dynamics toward discrete attractors as a shared source for both stimulus-specific and decision-consistent biases.
• Demonstrated that decisions dynamically steer memory states, influencing bias magnitudes.
• Neuroimaging and behavioral data confirmed the co-evolution of biases predicted by the attractor dynamics model.

Conclusions:

• Working memory representations are not static but exhibit dynamic drift toward attractors, explaining common perceptual biases.
• Decision-making processes actively modulate these memory dynamics, leading to decision-consistent biases.
• Recurrent neural network models provide a plausible neural substrate for how categorical decisions emerge from continuous sensory WM while incorporating these biases.