Advances in Approximate Bayesian Inference | Tweets by @MatthiasSBauer @adjiboussodieng @liyzhen2 @dawen_liang @s_mandt
Global Optimum
Joined September 2020
- Tweets43
- Following16
- Followers722
- Likes33
AABI retweeted
ABC is an exact algorithm, targeting a generalised Bayesian posterior arxiv.org/abs/2011.08644
Our new work, accepted at @aabi_org, explores how this view can help us with robustness to misspecification. Joint work with the brilliant @SebastianSchmon & @LauchLab.
AABI retweeted
Join us next week for our last session of AABI.
In the final #AABI2020 session on Wednesday, Feb 17, Jascha Sohl-Dickstein will present "Infinite Width Bayesian Neural Networks". Register at us02web.zoom.us/webinar/regi… or watch our Youtube Livestream at youtube.com/channel/UCZLWLfZ….
In the final #AABI2020 session on Wednesday, Feb 17, Jascha Sohl-Dickstein will present "Infinite Width Bayesian Neural Networks". Register at us02web.zoom.us/webinar/regi… or watch our Youtube Livestream at youtube.com/channel/UCZLWLfZ….
We also have two exciting contributed talks: "Roundoff Error in Metropolis-Hastings Accept-Reject Steps" by Matt Hoffman and "VIB is Half Bayes" by Alex Alemi. See approximateinference.org/sch… for details.
In the fifth #AABI2020 session on Wednesday, Feb 10, Mijung Park will present "ABCDP: Approximate Bayesian Computation & Differential Privacy". Register at us02web.zoom.us/webinar/regi… or watch our Youtube Livestream at youtube.com/channel/UCZLWLfZ….
In addition we will also have two contributed talks: "Conjugate Energy-Based Models" by Hao Wu and "Combining Pseudo-Point and State Space Approximations for Sum-Separable Gaussian Processes" by William Tebbutt. See approximateinference.org/sch… for details.
AABI retweeted
Come check out the work we did with Nikolai on how one can estimate the KL divergence (or the evidence if you're lucky) with gradient flows when there is no formula available! Looking at you Stein Variational Gradient Descent 👀
In addition we will also have two contributed talks: "Bijective-Contrastive Estimation" by Jae Hyun Lim and "Evidence Estimation by Kullback-Leibler Integration for Flow-Based Methods" by Nikolai Zaki. See approximateinference.org/sch… for details.
AABI retweeted
Exciting new work on contrastive estimation of EBM using invertible maps, with @jaehyunlim0606.
We use bijections to “generate” contrasting distributions and solve the corresponding classification problems similar to NCE. There are some new tricks, and some new challenges.
1/7
In addition we will also have two contributed talks: "Bijective-Contrastive Estimation" by Jae Hyun Lim and "Evidence Estimation by Kullback-Leibler Integration for Flow-Based Methods" by Nikolai Zaki. See approximateinference.org/sch… for details.
AABI retweeted
Yet another great symposium on Approximate Bayesian Inference! 🍻
Hugely recommend for the die-hard #Bayesian and #DeepLearning folks 🙌
In the third #AABI2020 session on Wednesday, Jan 27, Justin Domke will present "Some Embarrassing Questions about Variational Inference". Register at us02web.zoom.us/webinar/regi… or watch our Youtube Livestream at youtube.com/channel/UCZLWLfZ…
In the fourth #AABI2020 session on Wednesday, Feb 3, Fredrik Lindsten will present "Sequential Monte Carlo for Approximate Bayesian Inference". Register at us02web.zoom.us/webinar/regi… or watch our Youtube Livestream at youtube.com/channel/UCZLWLfZ…
In addition we will also have two contributed talks: "Bijective-Contrastive Estimation" by Jae Hyun Lim and "Evidence Estimation by Kullback-Leibler Integration for Flow-Based Methods" by Nikolai Zaki. See approximateinference.org/sch… for details.
The recordings for the past sessions are also available online at approximateinference.org/sch…
AABI retweeted
Tune in today at 5PM GMT at #AABI2020 for some great talks by @JustinDomke and @emiel_hoogeboom, as well as our own presentation and Q&A session (see below)! 😇 @aabi_org
Zoom: us02web.zoom.us/webinar/regi…
YouTube Livestream: youtube.com/channel/UCZLWLfZ…
"Expressive yet Tractable Bayesian Deep Learning via Subnetwork Inference” (Contributed Talk)
@edaxberger, @eric_nalisnick, @JamesAllingham, @JaviAC7, @jmhernandez233
Video: youtube.com/watch?v=HqyhSK_2…
Paper: openreview.net/forum?id=WqMl…
(2/11)
In the third #AABI2020 session on Wednesday, Jan 27, Justin Domke will present "Some Embarrassing Questions about Variational Inference". Register at us02web.zoom.us/webinar/regi… or watch our Youtube Livestream at youtube.com/channel/UCZLWLfZ…
In addition we will also have two contributed talks: "Argmax Flows: Learning Categorical Distributions with Normalizing Flows" by Emiel Hoogeboom and "Expressive yet Tractable Bayesian Deep Learning via Subnetwork Inference" by Erik Daxberger. See approximateinference.org/sch…
AABI retweeted
An awesome invited talk on 'Variational Dynamic Mixtures' via Mixture Density Networks for RNNs.
Also, some autoregression & ELBOs thrown in with some great use cases!
arxiv.org/abs/2010.10403
Followed by contributed talks in #Gaussian, and Alpha KL Divergence
#AABI2020 #Bayesian
In the second #AABI2020 session on Wednesday, Jan 20, Maja Rudolph will present "Variational Dynamic Mixtures". Register at us02web.zoom.us/webinar/regi… or watch our Youtube Livestream at youtube.com/channel/UCZLWLfZ…
We had a great second session today with speaker Maja Rudolph. Her talk was centered around work she did in this paper arxiv.org/abs/2010.10403.
Thanks to the contributed talk speakers and everyone who tuned in! See you next week for the third session with Justin Domke 🎊
In the second #AABI2020 session on Wednesday, Jan 20, Maja Rudolph will present "Variational Dynamic Mixtures". Register at us02web.zoom.us/webinar/regi… or watch our Youtube Livestream at youtube.com/channel/UCZLWLfZ…