Roman Ring (@Inoryy) header
Roman Ring (@Inoryy) profile picture

[email protected] Research Engineer @DeepMind Large Scale Deep Learning Is All You Need? AlphaStar, Gopher, Flamingo, ???

Joined July 2011
Photo by Roman Ring (@Inoryy)Photo by Roman Ring (@Inoryy)Photo by Roman Ring (@Inoryy)Photo by Roman Ring (@Inoryy)Photo by Roman Ring (@Inoryy)Photo by Roman Ring (@Inoryy)Photo by Roman Ring (@Inoryy)Photo by Roman Ring (@Inoryy)Photo by Roman Ring (@Inoryy)Photo by Roman Ring (@Inoryy)
Roman Ring retweeted
It is great to see the excitement about 🦩! As shown in different examples during the last few days, interacting with 🦩 has been quite fun, unique and sometimes mind blowing. However,🦩 has clear limitations as detailed in this 🧡! 1/11
4
70
14
386
Roman Ring retweeted
The title and some of the content are too sensationalist for my taste. Furthermore, content is mostly a recap of the discussions from mid '19, with no mention of their more recent DotA 2 paper, which was both very detailed and released without much PR fanfare.
1
8
Roman Ring retweeted
Replying to @Inoryy @jachiam0
Thing is, it's only a failure if you give up and never try anything again. As long as you keep going it's just a setback. Eventually, you'll just "brute force" the success at *something*, while having a diverse set of experiences to draw from.
1
1
5
Roman Ring retweeted
Replying to @marksaroufim
Spaced repetition shines as a tool for memorization, not learning. I have over 10k cards in Anki that range from useful VIM motions I rarely use to common tricks for solving differential equations. I can't imagine trying to maintain all of this information without it.
1
3
There has never been more a more exciting time to be a compiler nerd who is into machine learning.
5
11
2
152
Roman Ring retweeted
Really exciting news from the #MLIR team today: blog.google/technology/ai/ml… I'm so glad to be working with @clattner_llvm et al on building amazing OSS infra for ML and beyond!
1
14
49
job interview question: how would you build your own neural network from scratch? actual job requirement: the self-control to not write your own neural network from scratch
4
14
2
195
β€œWe hope that OpenSpiel could have a similar effect on general reinforcement learning in games as the Atari Learning Environment has had on single-agent reinforcement learning.” More info here: deepmind.com/research/open-s… venturebeat.com/2019/08/27/d…
51
143
Roman Ring retweeted
This is an excellent non-mathematical overview of how deep learning can be understood from the perspective of topology, information theory, causality, statistical physics, computation and neuroscience. Helps to see the big picture behind deep learning. arxiv.org/abs/1908.10206
26
89
Best debugger ever made is a good night's sleep
13
397
20
1,520
Excited that AlphaStar will be joining Battle.net for scientific evaluation & for a limited time! The updated agent: -Plays as and versus all races -Plays multiple ladder maps -Uses a camera interface and more restricted APMs FAQ: starcraft2.com/en-us/news/22… GL & HF!
StarCraft update! As part of ongoing scientific research into artificial intelligence, players can now opt-in for the chance to be matched against experimental versions of DeepMind's StarCraft II agent, AlphaStar, on the competitive ladder. Learn more: blizz.ly/DeepmindLadder
16
138
10
538
Roman Ring retweeted
Our first step toward creating AI agents with theory of mind: deep reinforcement learning agents can learn to take the perspective of another agent. This work was inspired by experiments done with chimpanzees, so we have monkeys and bananas on figures! arxiv.org/abs/1907.01851
1
10
42
status update: @DeepMindAI is beautiful they have endless coffee and better snacks everyone is asking @TensorFlow questions (GOOD ONES) I am never leaving – at London, England
7
10
2
191
110 patients seen this week. My reflection : life is hard for everyone. Behind closed doors, nobody has all their s**t together. Illness comes from physical, emotional, mental & social pain. Go easy on each other & go easy on yourself. I hope you have a restful #weekend x
16
90
2
562
Most interesting real-world problems are non-convex optimization problems. Theoretical CS taught me to fear and simplify these kinds of problems because "solving" them is very hard. Machine learning taught me to embrace them: good solution today > perfect solution never.
9
47
3
409
Math is beautiful. Our education system does not often reveal this. It's a lost opportunity to inspire creative young minds who are captivated by the incredible mysteries of our universe.
35
234
20
1,301
πŸ€– Day #51 (and winner #4 of @TensorFlow 2.0 socks! @Inoryy 🧦) Deep #ReinforcementLearning with TF 2.0: implementing an advantage actor-critic agent, solving classic CartPole-v0 environment. blog: inoryy.com/post/tensorflow2-… πŸ‘©β€πŸ’»: github.com/inoryy/tensorflow… πŸ“’: colab.research.google.com/dr…
1
3
16
Roman Ring retweeted
The Count of Monte Carlo: a wrongly imprisoned man sets to exacting elaborate revenge on random people, expecting that he'll eventually get the right people if he does so for long enough.
23
277
13
1,179
Roman Ring retweeted
ALL tech alarmism needs a pseudo-science philosopher (unlfasifiable) + a promoter A.I (2018 Bostrom + Musk) Nano (2000: Drexler + Joy) But smartest & most successful in actual AI agree w/ @sapinker+@kevin2kelly+@rodneyabrooks It’s nonsense––here’s WHY venturebeat.com/2018/12/17/g…
7
24
4
67
πŸ›Έ #TFadvent #15: Reaver, by @Inoryy: a @StarCraft II Deep Reinforcement Learning Agent, following in @DeepMindAI's footsteps. Works with Gym, #Atari, and MuJoCo, as well! πŸ’» Code: github.com/Inoryy/reaver-pys… πŸ“’ @GoogleColab: colab.research.google.com/dr… πŸ“½οΈ Demo: youtube.com/watch?v=gEyBzcPU…
1
6
20