Gallant lab/UC Berkeley
A Unified Framework for Receptive Field Estimation
Tuesday 07th of March 2006 at 05:00pm
In an attempt to characterize the functional implementation of sensory neurons, many algorithms have been used to estimate the neuron’s stimulus-response mapping function. These include spike-triggered average, normalized reverse correlation, linearized reverse correlation, ridge regression, regularized linear regression, spike-triggered covariance, neural networks, maximal informative dimensions, kernel regression, and boosting. These algorithms seem very different and make very different assumptions about the data and the system under investigation. Despite their name and origin, most of them are actually very similar when viewed as a maximum a posteriori (MAP) estimate. We will describe this framework for mapping function estimation, and show how the above algorithms fit into the framework. Under this unifying framework, the assumptions built into each of the existing algorithms are revealed and made explicit as priors, noise distributions and model class. Hence, the investigators could choose a more appropriate algorithm base on the assumptions they are willing to make. Moreover, this framework can facilitate the development of novel algorithms that incorporate biophysically plausible assumptions.
Join Email List
You can subscribe to our weekly seminar email list by sending an email to
email@example.com that contains the words
subscribe redwood in the body of the message.
(Note: The subject line can be arbitrary and will be ignored)