A SkillsCast for this session is not available.
https://github.com/deepmind/spiral
This paper proposes a very interesting application of RL on generating images. Current methods that combine deep learning and renderers are limited by hand-crafted likelihood or distance functions, which creates a need for large amounts of supervision, or difficulties in scaling their inference algorithms to richer datasets.
To mitigate these issues, SPIRAL is an adversarially trained agent that generates a program which is executed by a graphics engine to interpret and sample images. You can see it performing here: https://www.youtube.com/watch?v=iSyvwAwa7vk&feature=youtu.be
YOU MAY ALSO LIKE:
Synthesizing Programs for Images using Reinforced Adversarial Learning
Ali Chaudhry
Ali is a self-taught programmer, currently doing a PhD in Artificial Intelligence and Education at University College London. His research focuses on applying human-centered learning science theories on Deep Reinforcement Learning algorithms. These days he's obsessed with GANs and Variational Autoencoders.