Related Experiment Videos
A neural network model of flexible spatial updating
Robert L White1, Lawrence H Snyder
1Department of Anatomy and Neurobiology, Washington University School of Medicine, St. Louis, Missouri 63110, USA.
Journal of Neurophysiology
|December 12, 2003
Summary
A neural network model shows how the brain updates spatial memory during gaze shifts. It suggests the brain prioritizes gaze velocity over position for accurate world-fixed target tracking.
Area of Science:
- Neuroscience
- Computational Neuroscience
- Cognitive Science
Background:
- Visuospatial processing relies on retinotopic coordinates in the cortex.
- Updating spatial representations is crucial during gaze shifts for both world-fixed and gaze-fixed targets.
Purpose of the Study:
- To model how recurrent neural networks (RNNs) store and update spatial information.
- To investigate flexible updating based on contextual cues and reference frames.
- To understand the neural mechanisms underlying spatial updating in the brain.
Main Methods:
- Trained a 3-layer RNN to simulate spatial memory storage and updating.
- Incorporated gaze perturbation signals and contextual cues for flexible updating.
- Analyzed network output for accuracy and preference for specific sensory signals (gaze position vs. velocity).
Main Results:
- The RNN accurately read out target positions when cued to either reference frame.
- Updating accuracy was reduced compared to static representation.
- The network preferentially used gaze velocity signals for updating world-fixed targets.
- Gaze position gain fields were absent when velocity signals were available.
Conclusions:
- RNN models can replicate behavioral patterns observed in spatial updating tasks.
- Gaze velocity signals play a critical role in updating world-fixed spatial representations.
- The findings offer insights into the neural computations supporting spatial memory and eye movements.