Haau-Sing Li 李 效丞 (@LHaausing) header
Haau-Sing Li 李 效丞 (@LHaausing) profile picture
Darmstadt 🇩🇪 & Lisbon 🇵🇹
Joined January 2020
Haau-Sing Li 李 效丞 retweeted
🚀 New blog: ModularRSI: Toward Generalizable Harness RSI Can agents truly improve their harnesses—or just overfit to benchmarks? We study Harness RSI through one lens: generalization beyond evolution experience. 🧵 Blog: recursive-self-improvement.n…
1
8
2
25
2,022
Haau-Sing Li 李 效丞 retweeted
LoopCoder-v2 is out 🚀 Loop Transformers reuse the same block for recurrent hidden-state refinement — letting models “think” more without simply stacking more layers. We study how many loops are actually worth it in Parallel Loop Transformers. 📜huggingface.co/papers/2606.1…
2
12
2
45
4,874
Haau-Sing Li 李 效丞 retweeted
2) We have a spotlight poster in the main conference, “Reranking Laws for Language Generation: A Communication-Theoretic Perspective” with @tozefarinhas and @LHaausing (Thursday Dec 12 11:00-14:00).
1
3
8
344
Haau-Sing Li 李 效丞 retweeted
Meet our fellow researchers representing @UKPLab at this year's @ACLmeeting: @IGurevych, @QianRuan_, @J_st_s, @androneil54, @trumancfy, Sheng Lu, @haishuofang, @Furkansahinuc, @KexinWang2049, @LHaausing, @AndreasWaldis, and a very special guest from @BathNLP, @Harish! #ACL2024NLP
9
1
41
3,058
Haau-Sing Li 李 效丞 retweeted
Please drop by our poster tomorrow and talk to my co-authors about how IVON can directly optimize a variational objective with similar performance and cost as Adam but many things on top (uncertainty, model adaptation, diagnosis, ...) Thu. 13:30-15:00, hall c 4-9, poster #1402
🤔 Variational learning is often thought to be impractical 🔥 Plot twist: it actually works better than Adam! Meet IVON, a new optimizer that brings the best out of variational learning – 🧵 (1/9) #NLProc #ICML2024 📰 arxiv.org/abs/2402.17641 youtu.be/TRNYnRRJBRg
1
7
22
2,003
Haau-Sing Li 李 效丞 retweeted
We need NLG models that can indicate when they are wrong and can handle language variation. A good representation of uncertainty is crucial. But where does uncertainty come from? How to represent it? How does it power NLG models? We unpack this arxiv.org/pdf/2307.15703.pdf 🧵👇1/7
5
37
4
141
24,874
Haau-Sing Li 李 效丞 retweeted
Don't miss »Python Code Generation by Asking Clarification Questions« by @LHaausing, @MohsenMes, @andre_t_martins and @IGurevych at 11:00 (17:00 CET) at Pier 7&8! arxiv.org/abs/2212.09885 #ACL2023NLP
1
5
485
Haau-Sing Li 李 效丞 retweeted
The monitors team of LxMLS2022! (plus @MarianaSCAlmei2 from org.)
2
7
41
Haau-Sing Li 李 效丞 retweeted
War has never been a solution & never will be Unbelievable such a war in #Ukraine is still possible in 21 century, in the middle of #Europe We are one family.This is an attack not by people but by politicians against #values #democracy & life Unite for a better world #NotoWar
12
88
Haau-Sing Li 李 效丞 retweeted
An exciting discission on "large language models" with @LHaausing and related topics on #codex #openai #copilot #gpt3 and his music journey in reggae genre. Plug in for next episode of #papseries open.spotify.com/episode/5gR… #thoughtsofaphd
1
3
Haau-Sing Li 李 效丞 retweeted
UKP is pleased to announce the second central call for applications for the ELLIS PhD Program. Want to earn your PhD while working with the best talent in machine learning in Europe? Take a look: ellis.eu/news/ellis-phd-prog…
10
34
Haau-Sing Li 李 效丞 retweeted
🚨 NEW PAPER ALERT 🤓 CURVES📈📈📈 ALERT 🚨 Transformer LMs pretrained on billions of words have dominated NLP, but which skills/features really depend on this huge scale? How much can models learn from more modest amounts of data? [1/10]
1
2
7
Haau-Sing Li 李 效丞 retweeted
🚨 NEW PAPER ALERT 🤓 CURVES📈📈📈 ALERT 🚨 Transformer LMs pretrained on billions of words have dominated NLP, but which skills/features really depend on this huge scale? How much can models learn from more modest amounts of data? [1/10]
4
49
3
222
Haau-Sing Li 李 效丞 retweeted
🔥EMNLP PAPER ALERT🔥 Pretrained models can learn to represent linguistic features, but do they learn that they should actually *use* those features? And if so, how long does it take? [1/11]
1
55
5
265