Camera ready for the Thermodynamic Variational Objective is up! It includes additional experiments, an expanded appendix, and a link to the code. arxiv.org/abs/1907.00031 Tuan Anh Le Frank Wood PLAI Group

1/ New on arXiv: "Amortized Population Gibbs Samplers with Neural Sufficient Statistics" arxiv.org/abs/1911.01382. Work by: Hao Wu (Hao Wu), Heiko Zimmermann (Heiko Zimmermann 🦋 [email protected]), Eli Sennesh (Eli Sennesh), and Tuan Anh Le (Tuan Anh Le). (thread below)

#CallForSubmissions #CallforPapers PROBPROG 2020 is seeking Extended Abstracts & Syndicated Submissions by *Jan 10 2020* for our April 2020 international conference dedicated to probabilistic programming! probprog.cc/#call

Check out our extensive review paper on normalizing flows! This paper is the product of years of thinking about flows: it contains everything we know about them, and many new insights. With @eric_nalisnick, DaniloJRezendeAliasAccount, Shakir Mohamed, Balaji Lakshminarayanan. arxiv.org/abs/1912.02762 Thread 👇

Excited to present the Thermodynamic Variational Objective at #NeurIPS2019! Come say hi to Frank Wood, Tuan Anh Le and myself :) East Exhibition Hall B + C #194 Wednesday at 5:00pm vmasrani.github.io/assets/neurips…

Our NeurIPS paper "Efficient Probabilistic Inference in the Quest for Physics Beyond the Standard Model" will be in the 5:30pm poster session, East Exhibition Hall B + C #155 #NeurIPS2019

Hey Neurips: come to our paper’s poster to learn all about multimodal-VAE! Our session is tomorrow (Wednesday) 5pm-7pm, No.150 at east exhibition hall. #NeurIPS2019 Paper: arxiv.org/abs/1911.03393 Code: github.com/iffsid/mmvae

New ELLIS unit brings together #AI experts from Engineering Science, Oxford Oxford Comp Sci Oxford Statistics, to shape how machine learning and artificial intelligence will change the world. eng.ox.ac.uk/news/new-oxfor…

We are delighted to announce creation of an ELLIS unit at Oxford spanning Engineering Science, Oxford Oxford Comp Sci Oxford Statistics . OxCSML faculty Chris Holmes and Yee Whye Teh are also helping co-direct the Robust ML programme in ELLIS.

Our tutorial on importance sampling and sequential Monte Carlo methods is now published in Foundations and Trends in ML. Should be available at NeurIPS. Highlights include learning proposals, target distributions, unbiasedness results and more. nowpublishers.com/article/Detail…

They say the field moves fast, but goodness... Rob Brekelmans is presenting a follow-up paper on the TVO this Friday at the #Neurips2019 workshop on Information Theory and Machine Learning: sites.google.com/view/itml19 #neurips2019

Tomorrow at ICML: Amortized Population Gibbs Samplers with Neural Sufficient Statistics Poster: icml.cc/virtual/2020/p… arXiv: arxiv.org/abs/1911.01382 Work by Hao Wu (Hao Wu), Heiko Zimmermann (Heiko Zimmermann 🦋 [email protected]), Eli Sennesh (Eli Sennesh), and Tuan Anh Le (Tuan Anh Le) [thread]

Hao Wu did a truly amazing job in this project!

When learning VAEs, is it possible to get a good signal-to-noise ratio for the importance weighted bound without reparameterization? Yes, and it improves the learning of discrete VAEs! Valentin Liévin Andrea Dittadi Paper arxiv.org/pdf/2008.01998… Github github.com/vlievin/ovis

When looking at a scene, we automatically parse what is where and readily understand the relationships between any objects in the scene. How can we represent and learn models that support such compositional, symbolic reasoning – while permitting efficient inference?

Humans can learn general-purpose and meaningful representations from perception by leveraging suitable abstractions. For example, being able to write out numbers involves the generic ability to reason about what strokes to use… [1/6]

Our work on Drawing out of Distribution (DooD) will be presented at #NeurIPS2022! See you there 🙂. CoCoSci MIT Tuan Anh Le ExLab

Happy to share our new paper in nature on using Deep Reinforcement Learning for discovering new, faster algorithms to multiply matrices: dpmd.ai/nature-alpha-t… #AlphaTensor More in thread👇(1/n)

Hanging with my friends and best paper award winner Rob Brekelmans at #ICML2024