Decomposing Classification For Mixture Models

So I recently learned about the ability to convert IPython notebooks into slides using nbconvert. Not sure how I overlooked that capability of nbconvert. There are some (somewhat out of date) references on Damián Avila's blog.

I'm pretty excited about the idea of using IPython notebooks as a more effective way of communicating results within the LHC collaborations. As part of the DIANA/HEP project, I'm trying to promote these kinds of collaborative tools that improve reproducibility and streamline (internal or external) sharing. There are some obstacles because most of the 200,000 presentations / year at the LHC are internal documents not meant for public consumption. This means either integrating nbconvert directly into indico or providing a similar service as nbviewer that can deal with CERN authentication and credentials.

Anyways, since I'm learning about this, I thought I'd try it out by writing a few slides related to Section 5.4 of my recent paper Approximating Likelihood Ratios with Calibrated Discriminative Classifiers, http://arxiv.org/abs/1506.02169. The slides are just a demonstration of some basic algebra, but I've recently started working with Juan Pavez on demonstraiting this technique with real classifiers (scikit-learn, theano, TMVA, ...). Hopefully more to come on this topic.

I've embed the slides below, or if you are on your phone you can try this direct link to the slides since I'm not sure how to do a responsive embed. Click on the slides once and then you can use your \(\leftarrow\) \(\uparrow\) \(\rightarrow\) \(\downarrow\) keys to navigate (or click the navigation in the bottom right of the slides).