FW: [columbia-ml-reading] last minute: durk kingma speaks tomorrow (wednesday) at 10am

[Dawen Liang]

Last minute notice... It should be of interests to many people in this group.

Best,

Dawen

---------- Forwarded message ----------
From: Alp Kucukelbir <akucu...@gmail.com>
Date: Tue, Oct 20, 2015 at 3:30 PM
Subject: [columbia-ml-reading] last minute: durk kingma speaks tomorrow (wednesday) at 10am
To: "columbia-...@googlegroups.com" <columbia-...@googlegroups.com>, nycml...@cs.nyu.edu, machine_lear...@cs.nyu.edu, ml-stat-ta...@lists.cs.princeton.edu


Dear all,

Durk Kingma, who is a PhD student with Max Welling at the University
of Amsterdam is visiting Columbia. He has graciously agreed to give a
last moment talk tomorrow. Durk has made some impressive contributions
to the field of stochastic gradient based variational inference.
Should be a very interesting talk!

Details are below.

Time: Wednesday, 10am

Location: 750 CESPR at Columbia University

Title: Efficient Inference and Learning with Intractable Posteriors?
Yes, Please.

Abstract:

We discuss a number of recent advances in Stochastic Gradient
Variational Inference (SGVI).

- Blending ideas from variational inference, deep learning and
stochastic optimization, we derive an algorithm for efficient
gradient-based inference and learning with intractable posteriors.

- Applied to deep latent-variable models with neural networks as
components, this results in the Variational Auto-Encoder (VAE), a
principled Bayesian auto-encoder. We show that VAEs can be useful for
semi-supervised learning and analogic reasoning.

- Further improvements are realized through a new variational bound
with auxiliary variables. Markov Chain Monte Carlo (MCMC) can be cast
as variational inference with auxiliary variables; this interpretation
allows principled optimization of MCMC parameters to greatly improve
MCMC efficiency.

- When applying SGVI to global parameters, we show how an order of
magnitude of variance reduction can be achieved through local
reparameterization while retaining parallelizability. Gaussian Dropout
can be cast as a special case of such SGVI with a scale-free prior.
This variational interpretation of dropout allows for simple
optimization of dropout rates.

Cheers,
Alp

--
All info for upcoming presenters, reading material, locations, and times is in this Google sheet (p.s., please sign up to present):
https://docs.google.com/spreadsheets/d/1e_2pD0Hb9neSX-8Pf0sbHS4KtdLs-X_Pg9AKg86hz0U/edit?usp=sharing
---
You received this message because you are subscribed to the Google Groups "columbia-ml-reading" group.
To unsubscribe from this group and stop receiving emails from it, send an email to columbia-ml-reading+unsub...@googlegroups.com.
To post to this group, send email to columbia-ml-reading@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/columbia-ml-reading/CAL2cgJYB9SZXG7GK%3DDTkLQkewa4BVS-%3DQ5oJ79ZZJ9ckg%3D86Ug%40mail.gmail.com.
For more options, visit https://groups.google.com/d/optout.