New #TMLR-Paper-with-Video:

A Multi-Fidelity Control Variate Approach for Policy Gradient Estimation

Xinjie Liu, Cyrus Neary, Kushagra Gupta et al.

https://tmlr.infinite-conf.org/paper_pages/zAo0L7Dcqt

#reinforcement #reinforce #trained

2/By training ants with local vision and path context via #backpropagation and #REINFORCE, NAnts can collaboratively paint complex shapes like geckos, butterflies, and snails.

Even if partially erased or started anywhere on the grid, they dynamically repair the image! 🦎✨