This paper provides an overview of parameter-searching methods for single-neuron compartmental neural models. We compare several different parameter-search methods on a variety of neural models. The methods including conjugate-gradient descent, stochastic search, genetic algorithm, and simulated annealing. The models include a very simple one-compartment model with four active conductances, two passive dendritic models, and a moderately realistic model of a layer 2 pyramidal cell in piriform cortex. We discuss the relative advantages and disadvantages of the different methods for the different classes of problems and make specific suggestions on how to optimally go about the parameter searching process with neural models.
This paper provides a tutorial on Bayesian methods from the perspective of neuroscientists (specifically compartmental neural modelers), focusing on the problem of comparing two different compartmental neural models or model classes. We use Bayesian methods to estimate the relative probabilities of the models given an agreed-upon data set. We use two different methods to estimate the likelihoods: one based on a simple sum-of-squared differences in spike times (the Gaussian ISI model), and the other using a variable-rate Poisson model. We also illustrate the use of Bayesian methods in the context of a realistic model of a layer 2 superficial pyramidal cell from olfactory (piriform) cortex tuned on real neural data. In the latter case, we use Bayesian methods to determine the optimal level of noise added to the model so that the variability in the spike times in the model is optimal given the data. We also include a brief discussion of the theoretical foundations of Bayesian inference.
| Go back to my home page. | Last updated April 11, 2017 |
Mike Vanier (mvanier@cs.caltech.edu)