Alex Zhang
@a1zhang
Followers
10,698
Following
218
Media
3
Statuses
42
currently @snap research, previously @apple , @princeton ‘24 | interested in nlp/rl research & efficient ML systems
USA
Joined December 2015
Don't wanna be here?
Send us removal request.
Explore trending content on Musk Viewer
TEMU
• 446492 Tweets
最大5,000円報酬
• 236820 Tweets
Semarang
• 215422 Tweets
روما
• 203167 Tweets
領空侵犯
• 155196 Tweets
Arlington
• 124729 Tweets
Sven
• 123805 Tweets
Finale
• 34051 Tweets
Scout
• 31569 Tweets
#庭ラジ
• 28840 Tweets
Switchの後継機種
• 24769 Tweets
ソフトメーカーラインナップ
• 24381 Tweets
Rayo Vallecano
• 22269 Tweets
Medina
• 21958 Tweets
D-1 TO CHANYEOL SOLO
• 19199 Tweets
伊藤さん
• 17643 Tweets
瑞希誕生日
• 12657 Tweets
Pinned Tweet
for the past 2 weeks I’ve been reading every abstract from
#neurips2023
here are my notes about what I discovered:
i think everyone can learn something new from this, and I hope this resource is useful!
7
83
466
i can now say Princeton CS thought i was their biggest nerd this year because they gave me an award and $5k for it (i’m still jobless tho lol)
15
2
192
ok so mathjax was (and still is) being pesky, so i had to reupload
to repeat: i made a super concise, annotated code guide for kolmogorov-arnold nets (which r scary but not rly)! focuses less on b-splines and more on researching them; plz take a read!❤️
2
3
20
Very happy to share this work that we’ve been doing for the past year (also was part of my junior paper)!
At a high-level, we want to be able to modulate the behavior of AI agents through language. In our work, we do this by learning WMs that can read text.
📢 Excited to announce our new paper
Language-guided world models: A model-based approach to AI control
• We develop LWMs: world models that can read texts to capture new environment dynamics
• These models enable humans to efficiently control agents by providing language
4
46
203
0
1
14
spent some time tinkering w/ KANs by
@ZimingLiu11
and wrote up an *Annotated KAN w/ code* for ppl who want to quickly try them out (i’m unaffiliated btw, was just doing some research on them)
tried to make it super barebones and readable > efficient
1
0
11
Current video understanding systems do not work over long streams of video bc of a lack of labelled data.
So we created ASAP, a fully automated sports annotation pipeline to generate high-quality video annotations, and LCric, an 1000+ hour benchmark to go with it!
Excited to introduce ASAP, an automated pipeline for annotating sports matches! ⚽🏈🏀
Using ASAP, we created LCric, a fully-annotated 1000+ hour video understanding benchmark of full-length cricket matches. 🏏
Page:
Paper:
(1/5)👇
1
4
32
0
2
6
@thegautamkamath
Thank you for the kind words! I feel like this kind of exercise is especially useful if you are already aware of what’s popular so you pay special attention to novel ideas and more obscure research directions
1
0
7
tldr; i added custom masks to FlashAttention2 (Triton implementation) this weekend. it's easy to use & mostly functions the same as the original.
story is my friend
@AlperCanberk1
's team needed this feature bc xformer's memory_efficient_attention doesn't work well
3
0
6
@cocoweixu
funny story my original plan was to do this for iclr2024 instead but when I started the process they hadn’t finished reviews… unfortunately I go back to school in a week lol so it’ll have to be someone else
0
0
2
@imrahulmaddy
yeah I’ve been meaning to clean that up. I pulled a bunch of papers from arxiv originally before realizing the bibtex is on openreview, but I’m gonna spend some time changing them all to openreview (bc it includes conference as well) soon
1
0
1
also plz view in light mode, idk why but the code renderer (and MathJax sometimes) is really buggy on ai-folio
0
0
1
@saurishhh
it’s ok haha the dept actually only considers your best 8 departmentals so you can technically replace it
2
0
1
code & notebooks here:
wasn’t able to cook up any groundbreaking results, but i hope the examples are informative
also, would love feedback since i plan to formalize my notes for others to use in the future
1
0
1