Related Experiment Videos
Parameter estimation in a model for multidimensional recording of neuronal data: a Gibbsian approximation approach
L M Ould Mohamed Abdallahi1, C La Rota, M Béguin
1Xerox Research Centre Europe, 6 Chemin de Maupertuis, F-38240, Meylan, France.
Biological Cybernetics
|September 25, 2003
Summary
This study introduces advanced numerical methods for analyzing brain activity using spatiotemporal lattice models. These improved techniques enhance parameter estimation for cortical activity data, offering new insights into neural processing.
Area of Science:
- Computational neuroscience
- Statistical physics applied to neural systems
Background:
- Cortical activity analysis often employs spatiotemporal lattice models.
- Accurate parameter estimation is crucial for understanding neural dynamics.
Purpose of the Study:
- To propose and evaluate improved numerical procedures for parameter estimation in spatiotemporal lattice models.
- To enhance the analysis of cortical activities recorded from diode arrays.
Main Methods:
- Utilized approximations from statistical physics, including Gibbsian and mean-field approaches.
- Implemented pseudomaximum-likelihood estimators for parameter estimation.
- Evaluated estimator performance using Monte Carlo simulations.
Main Results:
- Demonstrated that mean-field approximations effectively reduce estimator variance for large diode arrays (144 diodes).
- Validated the improved numerical procedures through simulations.
Conclusions:
- The developed methods offer more reliable parameter estimation for cortical activity analysis.
- New interpretations of Guinea pig auditory cortex data in response to auditory stimuli were derived using these enhanced methods.