Excited to share @davide__belli's work (with minor contributions by myself) on the Generative Graph Transformer, to be presented at the Graph Representation Learning workshop at @NeurIPSConf.
— Thomas Kipf (@thomaskipf) November 1, 2019
Blog: https://t.co/jokhNVRsXU
Paper: https://t.co/fQTs60HUJ1
1/2 pic.twitter.com/zVbgCIzDcK
On the Cross-lingual Transferability of Monolingual Representations
— Thomas Scialom 😷 @emnlp2019 (@ThomasScialom) November 1, 2019
By @artetxem, @seb_ruder & @DaniYogatama
Interesting work on the capacity of models trained for a task in a specific language to transfer well in other languages.
New multilingual QA evaluation dataset
TL;DR👇
Prescribed Generative Adversarial Networks
— hardmaru 😷 (@hardmaru) November 1, 2019
Adding noise to the generator's output can help prevent mode collapse common in GANs, and also allows approximate log-likelihood evaluation. It's like killing two birds with one stone!
By @adjiboussodieng et al. https://t.co/q0bgXh3nTW pic.twitter.com/LIQwD9Hi99
We explore a simple approach to task-oriented dialog. A single neural network consumes conversation history and external knowledge as input and generates the next turn text response along with the action (when necessary) as output. Paper: https://t.co/EI9L4Qc5BD 1/4 pic.twitter.com/BjptMZBHvm
— Arvind Neelakantan (@arvind_io) November 1, 2019
SinGAN - Official pytorch implementation of the paper: "SinGAN: Learning a Generative Model from a Single Natural Image" https://t.co/O96ymf4N6o
— Python Trending (@pythontrending) October 31, 2019
GANs are powerful generative models but they suffer from mode collapse and are hard to evaluate on test data. We developed PresGANs to address these two limitations: https://t.co/dwVM7bDsiD pic.twitter.com/bJJQInCqjH
— DeepMind (@DeepMindAI) October 31, 2019
Our new @nature paper: AlphaStar is the first learning system to reach the top tier of a major esport without any game restrictions, achieving Grandmaster status in StarCraft II.Â
— DeepMind (@DeepMindAI) October 30, 2019
Researchers have been working on the StarCraft series for over 15 years. https://t.co/1vnbjIfboR pic.twitter.com/ohOotyrhB0
ICCV 19 - The state of (some) ethically questionable papers https://t.co/yUm6P7nChw
— /MachineLearning (@slashML) October 30, 2019
Facebook highlights AI that converts 2D objects into 3D shapes - Easy read for people interested in an overview of these computer vision efforts. https://t.co/JNCaZ52svL
— Nando de Freitas (@NandoDF) October 30, 2019
This is a wonderful paper that should be read by every ML researcher concerned with explainability. I dont know how it escape my attention. I would shorten it more, skip the "according to so and so.." and present ML folks with a computer-minded taxonomy of explanation #Bookofwhy https://t.co/UGvL1YYV0W
— Judea Pearl (@yudapearl) October 30, 2019
"BANANAS: Bayesian Optimization with Neural Architectures for Neural Architecture Search" -- there's definitely an opportunity to get a cool name out if it if you use a Bayesian strategy ... incl other goodies ;) https://t.co/9HiISEoSOl pic.twitter.com/JdOxNhkgaZ
— Sebastian Raschka (@rasbt) October 30, 2019
Summarization is one of the most important & least solved tasks in #NLProc
— Richard Socher (@RichardSocher) October 29, 2019
Problem with all #DeepLearning models: they are not optimized for factual correctness
We introduce a new task, dataset and model. Work by @iam_wkr @BMarcusMcCann @CaimingXiong
Paper https://t.co/NWer7tLWnj pic.twitter.com/pumDBw4RbO
