Category: GLMs

• SFN 2014 poster – converging encoding strategies in dorsal and ventral visual streams

  I have a poster session on Sunday afternoon at SFN 2014 in DC. It’s on a spiffy new method I’ve been working on for estimating the nonlinear transformation performed by an ensemble of sensory neurons, and its application to understanding visual representation in the dorsal and ventral visual streams. Some background: there’s a growing consensus […]

• PhD thesis – Parametric Models of Visual Cortex at Multiple Scales

  *Updated Sun April 6th* Well it’s done! I successfully defended my thesis on April 3rd. I now have what I’ve been longing for all these years – an obnoxious title that I can remind people of whenever I’m about to lose an argument. Unfortunately, this only works when erguing with non-PhDs. Entitled Parametric Models of […]

• Tips on using crcns datasets

  I’ve mentioned before that the CRCNS web site has a number of neural datasets available for download. To save you some time, here’s some tips to get you up and running for specific datasets. V2-1 data Jack Gallant’s V2 dataset is really interesting; I think it’s fair to say that we know very little about […]

• Why second-order methods can be futile in non-convex problems

  I’ve been working on fitting a convolutional model of neurons in primary and intermediate visual cortex. A non-convex optimization problem must be solved to estimate the parameters of the model. It has a form similar to: There are some shared weights to further complicate things, but the most salient features is that it’s a 3-layer […]

• Fitting a spline nonlinearity in a Poisson model

  I was talking to Jeremy Freeman at CSHL and he asked about an easy way to fit a spline nonlinearity in a Poisson regression model. Recall that with the canonical exponential nonlinearity, we have the following setup: And the negative log-likelihood is given by: Start by fitting w by maximum likelihood. Compute . Then you […]

• Using an L1 penalty with an arbitrary error function

  The L1 penalty, which corresponds to a Laplacian prior, encourages model parameters to be sparse. There’s plenty of solvers for the L1 penalized least-squares problem, . It’s harder to find methods for other error functions than the sum-of-squares. L1General by Mark Schmidt solves just such a problem. There’s more than a dozen different algorithms implemented […]