📚 My mini-book on amortized optimization is officially published! Via the Foundations and Trends® in Machine Learning journal (https://www.nowpublishers.com/MAL)

Buy a physical copy: https://a.co/d/4IlD3oo
Free online version: https://arxiv.org/abs/2202.00665
Source code: https://github.com/facebookresearch/amortized-optimization-tutorial

Need a new, easy-to-use RL algorithm?
(Which is essentially DQN but for continuous control Tasks!)

DecQN was accepted at #ICLR2023

https://twitter.com/markus_with_k/status/1585274003631087617?t=ZNEtDNKLiQgNyyLjEp830w&s=19

I finished my first "real" blog post. This is more work than expected, but also quite a lot of fun tbh.

https://trappmartin.github.io/website/post/2023_01_polya/

Happy to hear what you think. Suggestions how to improve it, or any other feedback. I am new to this, so please be kind. :)

'Cauchy–Schwarz Regularized Autoencoder', by Linh Tran, Maja Pantic, Marc Peter Deisenroth.

http://jmlr.org/papers/v23/21-0681.html

#autoencoders #autoencoder #generative

Named Tensor Notation (TMLR version, https://arxiv.org/abs/2102.13196)

A rigorous description, opinionated style guide, and gentle polemic for named tensors in math notation.

* Macros: https://ctan.org/tex-archive/macros/latex/contrib/namedtensor

Named Tensor Notation is an attempt to define a mathematical notation with named axes. The central conceit is that deep learning is not linear algebra. And that by using linear algebra we leave many technical details ambiguous to readers.

Boosting @gabrielpeyre ‘s toot on Stein’s lemma, which is a simple yet powerful tool and useful for many problems. I discovered it too late but I hope learn it sooner https://mastodon.social/@gabrielpeyre/109555859791380660

Also see our paper on extensions of it to various mixtures of exp-family https://arxiv.org/abs/1910.13398

Oldies but goldies: K. Weierstrass, Über die analytische Darstellbarkeit sogenannter willkürlicher Functionen einer reellen Veränderlichen, 1885. Proved that polynomials are dense in continuous functions on an interval. https://en.wikipedia.org/wiki/Stone%E2%80%93Weierstrass_theorem

@Boltwvan https://user-images.githubusercontent.com/707462/208504865-0efd3dea-6dc6-40ad-8aa5-7819d03ddf0f.gif 🙂

@michael_dennis Mostly 2/3-dimensional settings which still nicely captures surfaces in the world, although we've gone slightly beyond to a few tens of dimensions in some settings:

https://arxiv.org/abs/2106.10272
https://arxiv.org/abs/2207.04711

@Boltwvan It's OK with me! I also meant to post one starting with the Twitter logo instead of the Mastodon one.... Updating soon :)

@michael_dennis I've mostly looked at modeling distributions on manifolds, but I've been wanting to branch out into other topics in geometry as it's relatively new to me. Also the geometry/RL intersection seems quite nice and not something I have done much of :)

#introduction

I am a scientist at Meta AI in NYC and study machine learning and optimization, recently involving reinforcement learning, control, optimal transport, and geometry. On social media, I enjoy finding and boosting interesting content from the original authors on these topics

I made this small animation with my recent project on optimal transport that connects continuous structures in the world. The source code to reproduce this and other examples is online at https://github.com/facebookresearch/w2ot