Conditional Counting: Approximate Multinomial Probit Regression
Add to your list(s)
Download to your calendar using vCal
If you have a question about this talk, please contact Emli-Mari Nel.
There are many models for regression on a latent function from discrete data. An interesting case is the probit link function. It tends to only be used for binary classification, because the multi-class case involves an intractable integral.
I will present an approximate inference algorithm for probit inference from multinomial (count) data. It defines an expressive prior on discrete probabilities, allows inference directly from counts (instead of individual discrete variables), and returns an approximate Gaussian posterior. This makes it easy to combine with linear Gaussian and Gaussian process models, providing a predictive model for counts over kernel Hilbert spaces.
This talk is part of the Inference Group series.
This talk is included in these lists:
Note that ex-directory lists are not shown.
|