| Title | : | ML4Science Seminar: Isaac Tamblyn (National Research Council of Canada) |
| Duration | : | 01:01:31 |
| Viewed | : | 0 |
| Published | : | 06-05-2020 |
| Source | : | Youtube |
Keep it simple - Understanding the capabilities and limitations of modern AI for Science
In this lecture, I will give an overview of some of our recent work focused on using modern AI tools (e.g. sequence models [1], value-based reinforcement learning [2], and generative adversarial models [3]) to control and extract knowledge from simple physical spin models to understand and improve the underlying algorithms. These results can be used to understand the current capabilities and limitations of modern AI and deep learning within the context of scientific discovery and point the way toward the development of new algorithms tailor-made for scientific research.
[1]
arxiv.org/abs/2003.02647
[2]
arxiv.org/abs/2003.00011
[3]
arxiv.org/abs/2002.07055
SHARE TO YOUR FRIENDS
Scan me
