Thursday, May 07, 2020

Automating the search for entirely new curiosity algorithms

Recommendable! Meta learning is a promising approach to automatically find better algorithms!



" ... The researchers created a “meta-learning” algorithm that generated 52,000 exploration algorithms. They found that the top two were entirely new — seemingly too obvious or counterintuitive for a human to have proposed. ... Four machines searched over 10 hours to find the best algorithms. More than 99 percent were junk, but about a hundred were sensible, high-performing algorithms. Remarkably, the top 16 were both novel and useful, performing as well as, or better than, human-designed algorithms at a range of other virtual tasks, from landing a moon rover to raising a robotic arm and moving an ant-like robot in a physical simulation. ..."



Automating the search for entirely new “curiosity” algorithms | MIT News



This is the underlying paper:

https://arxiv.org/abs/2003.05325 to be presented at ICLR 2020

No comments: