We show that for the task of simulated robot wrestling, a meta-learning agent can learn to quickly defeat a stronger non-meta-learning agent, and also show that the meta-learning agent can adapt to physical malfunction.
We’ve found that self-play allows simulated AIs to discover physical skills like tackling, ducking, faking, kicking, catching, and diving for the ball, without explicitly designing an environment with these skills in mind. Self-play ensures that the en…
On October 19 2017, Ubuntu 17.10 will be released and as many of you know it packs lots of significant changes. I spend a week testing the Beta 2 and in this “last minute” review, I document some of the less obvious features/gotchas of Ubuntu 17.10. I …
Professor Naftali Tishby passed away in 2021. Hope the post can introduce his cool idea of information bottleneck to more people.
Recently I watched the talk “Information Theory in Deep Learning” by Prof Naftali Tishby and found it very in…
We’re releasing an algorithm which accounts for the fact that other agents are learning too, and discovers self-interested yet collaborative strategies like tit-for-tat in the iterated prisoner’s dilemma. This algorithm, Learning with Opponent-Learning…
The Datumbox v0.8.1 has been released! Download it now from Github or Maven Central Repository. What is new? The main focus of version 0.8.1 is to resolve various bugs, update the depedencies and improve the code architecture of the framework. Here are…
[Updated on 2018-09-30: thanks to Yoonju, we have this post translated in Korean!]
[Updated on 2019-04-18: this post is also available on arXiv.]
Generative adversarial network (GAN) has shown great results in many generative tasks to replicate the r…
We’re releasing two new OpenAI Baselines implementations: ACKTR and A2C. A2C is a synchronous, deterministic variant of Asynchronous Advantage Actor Critic (A3C) which we’ve found gives equal performance. ACKTR is a more sample-efficient reinforcement …