Some changes to the site:

  1. Acknowledge @WTFJHT in the site description above.
  2. Reduce the text size of the navigation sidebar to lower the chance of overlapping with the footer (still need a permanent solution, though).
  3. Include fallback texts for each tweet. It’s essentially for the site to be searchable (a feature that will be added later).

Today we have many GAN and reinforcement learning-related tweets. Personally I’m most impressed by @WSJGraphics’s NLP piece for its neat visualization.

Add AI generated sound to AI generated paintings

Adding AI generated sound to my AI generated landscape paintings with @qosmo_inc's "Imaginary Soundscapes" paper - it's really fun. pic.twitter.com/5gGOZGhPgI

— Robbie Barrat (@DrBeef_) May 21, 2018

Woah listen to this- their AI thinks that my AI generated nude portraits are animals of some sort; so it assigns them these *really* freaky animal sounds... pic.twitter.com/O6LxBRxMEq

— Robbie Barrat (@DrBeef_) May 21, 2018

The paper is here; while it's short, it's definitely one of the cooler and more succinct papers I've read recentlyhttps://t.co/TpdWm74EYx

— Robbie Barrat (@DrBeef_) May 21, 2018

GAN

Ramen dishes

Interpolating through the latent space of ramen dishes uses a GAN https://t.co/gkBLuDS4ap

— Ian Goodfellow (@goodfellow_ian) May 21, 2018

Google Driveにラーメン二郎GAN動画の元ファイルを置きました。 https://t.co/764Zd5Izyw #ラーメン二郎 #GAN

— Kenji Doi (@knjcode) May 21, 2018

Crypto.ai

GAN + Ethereum + Anime = https://t.co/QyUgvThvHb pic.twitter.com/HUsIilntml

— hardmaru (@hardmaru) May 22, 2018

This project by @_aixile @alanyttian et al. They presented an earlier work as a workshop paper at NIPS2017. https://t.co/dknlcPy4Om https://t.co/uGXTdytlXh

— hardmaru (@hardmaru) May 22, 2018

I also think the anime dataset really needs to be de-biased. I believe these projects are usually trained on the dataset graciously scraped by @gwern called Danbooru2017, and more often than not the collected art are anime drawings of female characters. https://t.co/bQGaA9R9iF

— hardmaru (@hardmaru) May 22, 2018

Reinforcement Learning

Reinforcement Learning - Markov Decision Process https://t.co/HGzQhMRT6a pic.twitter.com/i27HpvZZPI

— KDnuggets (@kdnuggets) May 21, 2018

NIPS2018 Pommerman Competition

NIPS2018 “Pommerman” Competition, now with fresh prizes 🤖💣 https://t.co/2IyhBuhtD0

— hardmaru (@hardmaru) May 21, 2018

#nips2018 Pommerman Competition Prizes (https://t.co/6ZJeqCgj8S):

1st: $4k USD + $6k GCE,
2nd: $2k USD + $4k GCE,
3rd: $1k USD + $2k GCE,
4th: $3k GCE.

Top 2 learning agents each get a Titan V GPU.

Let the games begin!

— Pommerman (@Pommerman) May 21, 2018

Reproducing papers

Reproducing research papers is a really good way to practice your ML Skills, but it’s not easy. Read Matthew Rahtz’s insights from his first go at reproducing a Deep Reinforcement Learning paper: https://t.co/mqOnPOBqr7

— Kaggle (@kaggle) May 21, 2018

This post about DeepRL reproducibility has caused much discussion in the past while. In my view, I don't think it should be this difficult to reproduce a paper, even partially. That fact that it was so painful might imply that there are fundamental issues w/ the proposed methods. https://t.co/kVmDHfYbda

— hardmaru (@hardmaru) May 21, 2018

Duckietown

Building a Simple Self-Driving Car Simulator https://t.co/FSgRHyRZLd

— Maxime Chevalier (@Love2Code) May 21, 2018

Duckietown: self-driving car simulator environments for OpenAI Gym. Created by @Love2Code et al.
repo: https://t.co/UV1XJe5qXK
blog: https://t.co/QV26yTTrEI pic.twitter.com/490ETJCFVH

— hardmaru (@hardmaru) May 21, 2018

NLP

Some of my thoughts on ai and the future of chat bots:https://t.co/a9Fm1AXFkY

— Richard (@RichardSocher) May 21, 2018

Do you write like a fourth-grader? Find out what kind of impression your words make using algorithms that score context, mood and equivalent grade-level. https://t.co/XKey7dp9ny via @WSJGraphics pic.twitter.com/Wqc6P4Ygn2

— WSJ Graphics (@WSJGraphics) May 21, 2018

Transfer Learning

Better and faster text classification models by transfer learning presented by Sebastian Ruder @seb_ruder from @insight_centre, co-authored by Jeremy Howard @jeremyphoward from @fastdotai #InsightDL2018 pic.twitter.com/MppK2VgUN8

— Xavier Giró-i-Nieto 🎗 (@DocXavi) May 21, 2018

Slides for my keynote this morning on Successes and Frontiers of Deep Learning are now online. Mainly a high-level overview of some important and exciting application areas + some recent work on semi-supervised / transfer learning. https://t.co/uVcLtbHSuT

— Sebastian Ruder (@seb_ruder) May 21, 2018

Visualization

How to win a trade war --> https://t.co/oEhGYpKzul

— FiveThirtyEight (@FiveThirtyEight) May 21, 2018

For those that haven't seen it, matplotlib 2.2 has a layout engine. Update now and set https://t.co/BYw43RVBY6 to True in your matplotlib.rc - and live happily ever after. https://t.co/B6qx3yrqCx

— Andreas Mueller (@amuellerml) May 21, 2018

Interesting simulation experiment: Throwing 1,000 needles to estimate Pi. #DataScience #dataviz

Read more about Buffon's needle here: https://t.co/CiQRAeM0qlhttps://t.co/jKQRDfMUzw pic.twitter.com/LHCPACeevN

— Randy Olson (@randal_olson) May 21, 2018

Beautiful and Powerful Correlation Tables in R https://t.co/SX3EwfxnxN #rstats #DataScience

— R-bloggers (@Rbloggers) May 21, 2018

PyCon 2018 Videos

PyCon 2018 Videos https://t.co/xDW3DEDV6u #python #pycon pic.twitter.com/s874SRpCZl

— Python Weekly (@PythonWeekly) May 21, 2018

Fast.ai

One of our students has started on a nice overview of the fastai libraryhttps://t.co/h4NADgWQqp pic.twitter.com/Vw0DFcYkMu

— Jeremy Howard (@jeremyphoward) May 21, 2018

Complete Guide to Parameter Tuning in XGBoost https://t.co/IGpkYkfsP9

— Jeremy Howard (@jeremyphoward) May 21, 2018

Miscellaneous

Meghan's friends introduced her to a prince. The only people I get introduced to are tech bros who say "I have the next big startup idea. It's like Uber but for dogs and it uses blockchain and VR. I need you to build everything and I'll pay you in equity" smh.

— Jesslyn 🇮🇩 (@jtannady) May 19, 2018

By age 31, you should have conquered the largest empire known to man and founded 20 cities, retirement experts say pic.twitter.com/J9gT5JP06s

— François Chollet (@fchollet) May 21, 2018

The value is not in abstract ideation, it is in correctly evaluating the potential of your various ideas, and successfully executing on the most powerful ones.

— François Chollet (@fchollet) May 21, 2018

@ceshine_en

Inpired by @WTFJHT