Perceptual Decision-Making: Biases in Post-Error Reaction Times Explained by Attractor Network Dynamics.

Area of Science:

• Neuroscience
• Computational Neuroscience
• Cognitive Science

Background:

• Perceptual decision-making is widely studied using evidence accumulation models.
• Attractor network models offer an alternative, but their predictions for continuous trials are less explored.
• Previous work showed attractor networks can model repetition biases in decision sequences.

Purpose of the Study:

• To investigate sequential effects in perceptual decision-making using an extended attractor network model.
• To understand how network dynamics influence reaction times and accuracy across trials.
• To determine if intrinsic network properties can explain subtle behavioral effects without explicit feedback.

Main Methods:

• Extended a biophysical competitive attractor network model (Wong and Wang, 2006) with post-decision inhibition.
• Analyzed the conditions for successful trial sequences, avoiding attractor trapping or memory loss.
• Studied the impact of nonlinear network dynamics on reaction times and performance.
• Compared model predictions with empirical findings on sequential effects in behavioral experiments.

Main Results:

• The extended attractor network model successfully performs sequences of decision trials.
• Model dynamics explain how reaction times and accuracy vary across trials.
• The network exhibits post-error slowing and post-error improvement in accuracy, matching behavioral data.
• These sequential effects emerge without any feedback on decision correctness.

Conclusions:

• Intrinsic nonlinear dynamics of attractor neural networks can account for sequential effects in perceptual decision-making.
• Post-error slowing and improved accuracy arise from the network's internal mechanisms, not external reinforcement.
• This provides a biophysically plausible explanation for subtle behavioral phenomena in decision tasks.