Pingsheng Li
pingshengli.bsky.social
Pingsheng Li
@pingshengli.bsky.social
CS PhD, Neuro Undergrad @mcgillu, @Mila_Quebec, LiNC Lab @tyrell_turing, Prev Neuro-X @EPFL, intern @GatsbyUCL, Neuro AI Convergence DL&RL.
Minimalist, proactive, and pragmatic. love it. 😉
November 15, 2025 at 6:12 AM
I had similar feelings... The piece of code for Adam is probably more impactful than the whole paper (from pragmatic pov)
May 16, 2025 at 2:44 PM
A basic example from coding theory: if you know the distribution or frequency of characters occurrence, you can spend less bits when coding them (Hoffman code). So better compression requires better understanding of underlying distribution.
May 4, 2025 at 8:28 AM
Exactly. I mean that's how evolution basically works. Only those who solve the tasks survive, even tho sometimes survivors' solutions are suboptimal. Biology conforms to reality. Not the other way.
April 29, 2025 at 2:50 PM
i think policy gradient might be a counterexample. You dont need all vars that matters to do policy grad. Reward is all u need.
March 22, 2025 at 7:29 AM
"We must gradient - we will gradient!"
March 21, 2025 at 4:56 PM
Assuming it's not faking results, I guess they might just want to attract interested collaborators? Experienced engineer can reverse engineer a product given very limited information. Personal opinions.
January 31, 2025 at 12:59 PM