Related Experiment Videos
A model that accounts for activity in primate frontal cortex during a delayed matching-to-sample task
S L Moody1, S P Wise, G di Pellegrino
1Cognitive Science Department, University of California, San Diego, La Jolla, California 92093-0515, USA.
Summary
A recurrent neural network model for spatial delayed matching-to-sample tasks (DMS) used distinct storage and comparator units. Some premotor and prefrontal cortex neurons showed similar storage and comparator functions, aiding in match detection.
Area of Science:
- Computational neuroscience
- Cognitive neuroscience
Background:
- The spatial delayed matching-to-sample (DMS) task requires remembering a location and identifying a match.
- Neural computations for DMS involve memory and comparison mechanisms.
Purpose of the Study:
- To develop and analyze a recurrent neural network model for the DMS task.
- To compare the model's neural activity with single-neuron recordings from primate cortex.
Main Methods:
- A fully recurrent neural network was optimized for the DMS task.
- The model employed distinct 'storage' and 'comparator' units in its hidden layer.
- Model activity was compared to neural recordings from premotor (PM) and prefrontal (PF) cortex.
Main Results:
- The model utilized directional tuning in both storage and comparator units.
- Many PM and PF neurons exhibited characteristics of storage units.
- Neurons in both PM and PF regions showed comparator-like activity and dual functionality.
- Some PF and PM neurons closely resembled the model's output signal.
Conclusions:
- Recurrent neural networks can effectively model DMS task performance using specialized neural units.
- Primate PM and PF cortices implement DMS-related computations using neurons with storage and comparator functions.
- Neural implementations in the brain show similarities but also differences compared to the model's output neuron.