Excited to share our recent work combining model-free and model-based RL methods for planning. Our approach drastically reduces the number of planning computations and we see emergent search patterns (depth-first and breadth-first) used during planning.
https://t.co/NZHxvihzKA
๐ This is our poster with @normantasfi and Miriam Capretz for the Preregistration Experiment at NeurIPS! If you are interested in transfer learning for RL, stay tuned for updates!
@jdcampolargo This is something Iโve thought a lot about and donโt think there is any one single cause. With that said, I think a large contributing factor is that as you get older more of your life is nailed down.
MEGATHREAD TIME: In 40 tweets I will describe 40 powerful concepts for understanding the world. Some are complex so forgive me for oversimplifying, but the main purpose is to incite curiosity. Okay, here we go:
@cgenco@Shpigford The iPad works great for me but most of my work falls into one of: code (research or server side), plotting, writing, or reading papers. I will admit that there is some clunkiness, eg. when you are doing heavy multitasking, but I found the setup to be quite productive.
@cgenco@Shpigford I usually ssh onto a server and use tmux+vim+git etc. I have ESC mapped to Caps lock (via Blink). When I need to port forward it lets you do that with the usual ssh flags so using other apps like Juno (ipython) is easy.