Ultimate following list to keep updated in artificial intelligence

Alexandr Honchar
Apr 11, 2018 · 9 min read
Still using arxiv as main knowledge source?

Everyone who is working with technologies knows the joy (and pain on the other hand) of rapid updates in the field. It doesn’t matter if you’re web developer who needs to learn new frameworks every couple of weeks, data scientist who has to know how to implement latest deep learning papers or financial manager who needs to know recent ways to manage assets better — in any case you have to read a lot and learn a lot to stay on the top. Artificial intelligence field is on fire today — hundreds of research papers, GitHub code releases, infrastructure updates are appearing every week. And it’s really difficult to know everything of that — from new Tensorflow Serving versioning updates to some another GAN architecture augmented with memory applied for graph data.

My personal opinion is that is impossible to know all novelties in AI unless you have a team of PhDs analyzing different fields and sources and bringing organized reports personalized for you. But it’s definitely possible to have birds eye view, see the whole picture and pick things that will be useful for you. For example you’re working on some computer vision project and most probably you’re already applying latest neural network architectures, use some good regularization algorithms and in-box gradient-based optimizing algorithms, but you can do much more than that. You can check codes released by Facebook, apply NLP-based losses from Einstein AI, use neuroevolution optimizers how Uber and OpenAI are doing now and check out more interesting regularization than L2 + Dropout. Maybe it won’t matter at all, but most probably it will bring your product to the next level in front of competitors, that are using pre-trained ResNets and are happy with their 85% of accuracy on some task, or worse, using some open API for that.

Or maybe you prefer Reddit?

In this article I want to present a more organized approach on how to read AI-related news on different levels to make it really useful and impactful. Enough intros, let’s start :)

Industrial research

I think everyone knows OpenAI, I recommend to check all their research at least on high level, they open really cool problems, release code and will be super useful for those who does reinforcement learning.

DeepMind does much more diverse research from neural network architectures and reinforcement learning to generative models and symbolic reasoning, definitely worth to follow to make your picture much more broad.

Uber is a dark horse in AI world, but they do really good research in time series, evolutional optimization and reinforcement learning as well. They also show good engineering practices in their blog.

Microsoft does much more fundamental applied research — they are discovering different interesting topics — NLP, optimization, neural program synthesis, biomedical data processing, robotics and much more.

IBM has a lot of nice academic papers with applied flavor — I recommend to PhD students to check out how they organize research. What I also personally like (same as in Uber) — it’s how they separate engineering and pure research in different branches.

Facebook releases a lot of very useful code for computer vision and NLP. Not so many mind blowing things as in OpenAI, but I’m sure you’ll always find something to apply in your projects.

If you’re working with speech — you simply must learn everything what Baidu does.

And… Google Research :) Very diverse, much more interesting things appearing in Google Brain and DeepMind, but still.

Academic research

Well, I couldn’t not mention Stanford here.

Berkeley lately is showing a lot of interesting things related to reinforcement learning.

MIT is like Stanford in a way — no specific emphasis.

And I also wanted to mention couple of European universities that are worth to follow:

If you like arxiv and academic papers there, you would like to switch to Arxiv-Sanity for… better user experience :)

Practitioners

Ferenc Huszár, machine learning researching in Twitter is writing very nice about mathematics and explains topics very well:

Another mathematical blog, a bit restricted to optimization:

Sebastian Ruder is rather popular NLP researcher, I don’t miss his articles:

The only English news aggregator I read is this Facebook group (I follow it just because of creator, voice recognition practitioner Arthur Chan and content is very useful):

For Russian speaking readers I would like to mention Alexey Levenchuk’s blog, where he is showing his “bird’s eye” view on AI infrastructure time to time and popular VK page, which is an aggregator in fact, but I cannot not mention it.

Twitter

World impact

How to deal with all of this?

For example, I check Twitter feed every day — I see what people in AI are talking about, what papers they’re reading, what they mention and simply press “heart” on something what I find interesting:

What I liked on Twitter to read later

Then I take couple of days per week (usually two) to sit couple of hours and read what I’ve saved on Twitter and what are the main updates in companies and universities blogs — I make my own noes, check papers, write down main ideas and study something in depth if I see how it will be helpful for me right now. Usually every week I see at least couple of released codes on Github that are worth to try or at least to check out (it can be implementations from twitter, OpenAI’s implementations or anything else) and I take couple of hours on my third day for some coding. If there is something I don’t know how to apply but I believe it may be useful to share with someone or simply interesting — I save it in a Google Docs file which I review and update monthly:

My doc fine for AI news in 2018

Let’s summarize things a bit:

  1. You can follow big players to understand general trends, see good code and practices
  2. You can follow some particular personalities to get insights in your fields or catch something what’s ignored in the “big world”
  3. You regularly save, review and try to apply or at least summarize things you’ve learned about
  4. You use public news aggregators on your own risk

Of course, you should mix this list with practitioners from your field, add more business people, or some specific researches, but the main idea is the above described system of catching high level trends, to see their successful applications and apply them in your business.

Good luck and stay tuned! Please, add people or sources who you think are worth to follow in comments!

P.S.
Follow me also in Facebook for AI articles that are too short for Medium, Instagram for personal stuff and Linkedin!

Alexandr Honchar

Written by

🇺🇦 🇮🇹 Developing AI that works. Consulting, giving public speeches and blogging. Contact me to collaborate alex.honchar@neurons-lab.com