Stochastic attractor models of visual working memory

Area of Science:

• Cognitive Neuroscience
• Computational Psychology

Background:

• Working memory models often use stochastic attractor dynamics to explain trial-to-trial response biases.
• Previous models successfully accounted for biases in multi-trial visual working memory tasks.

Purpose of the Study:

• To adapt stochastic attractor dynamics for multi-item working memory within a single trial.
• To evaluate a multi-item model's ability to explain behavioral data from continuous-report tasks.
• To elucidate the mechanism of swap errors in working memory.

Main Methods:

• Developed a multi-item working memory model where fixed points represent items.
• Analyzed behavioral data from continuous-report working memory tasks.
• Investigated memory trace diffusion and capture by attractors to model swap errors.

Main Results:

• The multi-item model provides a superior account of behavioral data compared to previous approaches.
• A mechanism for swap errors is proposed: memory traces diffusing and being captured by other item attractors.
• Swap-error curves reveal a continuous temporal evolution of this process.

Conclusions:

• Stochastic attractor dynamics, extended to multi-item representations, effectively models working memory.
• The model explains swap errors as a consequence of memory trace diffusion and attractor capture.
• Empirical memory performance reflects a stochastic process with error-correcting dynamics, not purely diffusive processes.