Explore tweets tagged as #Approximation
I’ll be giving a talk titled “Reinforcement Learning with Function Approximation—Done Right” at Purdue University (@PurdueECE ) this Tuesday. If the topic interests you, I’d be delighted to see you there—details are in the attached PDF.
Policy Iteration’s super-power—monotonic improvement + guaranteed convergence—vanishes under general function approximation. To bring them, we introduce Reliable Policy Iteration (RPI) : #ReinforcementLearning #RL.@EshwarSR @today_itself @DalalGal
0
2
16
🚨Past work shows: dropping just 0.1% of the data can change the conclusions of important studies. We show: Many approximations can fail to catch this. 📢Check out our new TMLR paper (w/ David Burt, @ShenRaphael , Tin Nguyen, and @ta_broderick ) 👇.
1
8
17
@Anthony_Bonato Constant seems to be misleading . 🐾 (Changed to Wolfram, desmos have a precision issue) Although log 8x is a good approximation according to Wolfram, but then again maybe we cannot see it with the available scale. 🐾
1
0
4
Day 271.For the MJ users: --sref 463853408.#Promptshare in ALT.Bookmark or look under highlights for more. The images below are made with the above #Sref. I do my best to write a prompt that recreates that .visual so everyone can participate but it is only an .approximation.
6
5
26