<html><body><div style="font-family: garamond,new york,times,serif; font-size: 12pt; color: #000000"><div data-marker="__QUOTED_TEXT__"><p style="margin-left: 50px;" data-mce-style="margin-left: 50px;">Rajesh Ranganath will present his FPO, "Black Box Variational Inference: Scalable, Generic Bayesian Computation and its Applications" on Friday, 10/6/2017 at 11:00 in CS 105.</p><p style="margin-left: 50px;" data-mce-style="margin-left: 50px;">The members of his committee are as follows: David Blei (Adviser); Examiners: Sanjeev Arora, Barbara Engelhardt, and David Blei; Non Examiners: David Blei and Peter Orbanz (Columbia University, Department of Statistics)</p><p style="margin-left: 50px;" data-mce-style="margin-left: 50px;">A copy of his thesis, is available in Room 310.</p><p style="margin-left: 50px;" data-mce-style="margin-left: 50px;"><br>Everyone is invited to attend his talk. The talk abstract follows below.</p><p style="margin-left: 50px;" data-mce-style="margin-left: 50px;">Probabilistic generative models are robust to noise, uncover unseen patterns, and make predictions<br>about the future. These models have been used successfully to solve problems in<br>neuroscience, astrophysics, genetics, and medicine. The main computational challenge is<br>computing the hidden structure given the data—posterior inference. For most models of<br>interest, computing the posterior distribution requires approximations like variational inference.<br>Variational inference transforms posterior inference into optimization. Classically,<br>this optimization problem was feasible to deploy in only a small fraction of models.<br>This thesis develops black box variational inference. Black box variational inference is<br>a variational inference algorithm that is easy to deploy on a broad class of models and has<br>already found use in models for neuroscience and health care. It makes new kinds of models<br>possible, ones that were too unruly for previous inference methods.<br>One set of models we develop is deep exponential families. Deep exponential families<br>uncover new kinds of hidden pattens while being predictive of future data. Many existing<br>models are deep exponential families. Black box variational inference makes it possible for<br>us to quickly study a broad range of deep exponential families with minimal added eort<br>for each new type of deep exponential family.</p><p style="margin-left: 50px;" data-mce-style="margin-left: 50px;">The ideas around black box variational inference also facilitate new kinds of variational<br>methods. First, we develop hierarchical variational models. Hierarchical variational models<br>improve the approximation quality of variational inference by building higher-fidelity<br>approximations from coarser ones. We show that they help with inference in deep exponential<br>families. Second, we introduce operator variational inference. Operator variational<br>inference delves into the possible distance measures that can be used for the variational optimization<br>problem. We show that this formulation categorizes various variational inference<br>methods and enables variational approximations without tractable densities.</p><p style="margin-left: 50px;" data-mce-style="margin-left: 50px;">By developing black box variational inference, we have opened doors to new models, better<br>posterior approximations, and new varieties of variational inference algorithms.</p></div></div></body></html>