Ziqian Lin @myhakureimu

LLM, ICL, ML, Recommendation myhakureimu.github.io Madison Joined July 2021

Tweets

59
Followers

170
Following

175
Likes

80

Kangwook Lee @Kangwook_Lee

4 months ago

Proudly presenting three doctors from my research group! 🤗 Congratulations 🥳 Dr. Yuchen Zeng @yzeng58 Dr. Ziqian Lin @myhakureimu Dr. Ying Fan @yingfan_bot + I will be posting highlights of their amazing research achievements soon.... stat tuned ... ;)

7 7 123 9K 5

Download Image

Andrew Lampinen @AndrewLampinen

4 months ago

How do language models generalize from information they learn in-context vs. via finetuning? We show that in-context learning can generalize more flexibly, illustrating key differences in the inductive biases of these modes of learning — and ways to improve finetuning. Thread: 1/

8 152 768 98K 689

Download Image

Qingyun Wang @eagle_hz

5 months ago

Thank you @hengjinlp so much for mentoring me since my junior year! It feels like yesterday that you provided detailed feedback and helped refine my submission for my very first paper in ACL 2018. To prospective students and interns: I'm currently recruiting passionate students…

Heng Ji @hengjinlp

5 months ago

0 2 68 11K 6

0 5 36 6K 5

Max Forbes @maxforbes

6 months ago

working on a post that's basically "how to get a paper accepted," using a case study one of my own that went from reject (2.5, 3, 3) to accept (4, 4.5, 4.5) with just one week of revisions

6 71 551 45K 735

Download Image

Kangwook Lee @Kangwook_Lee

6 months ago

1/ Super excited to share our new work “LLM-Lasso,” led by my collaborators from Stanford! tldr; We've reimagined the classic Lasso algorithm (by @robtibshirani), which uses ℓ1 regularization to select a sparse subset of features!

8 75 330 32K 224

Download Image

Core Francisco Park @corefpark

7 months ago

💥New Paper! Algorithmic Phases of In-Context Learning: We show that transformers learn a superposition of different algorithmic solutions depending on the data diversity, training time and context length! 1/n

7 64 433 37K 342

Download Image

Kangwook Lee @Kangwook_Lee

7 months ago

Happy to share our latest work on VersaPRM! github.com/UW-Madison-Lee… TL;DR: VersaPRM is the first fully open-source Process Reward Model (PRM), including data, code, and weights. It enhances LLM accuracy using test-time compute algorithms — extending beyond just mathematics!

2 22 114 10K 49

Download Image

clem 🤗 @ClementDelangue

9 months ago

Just 10 days after o1's public debut, we’re thrilled to unveil the open-source version of the groundbreaking technique behind its success: scaling test-time compute 🧠💡 By giving models more "time to think," LLaMA 1B outperforms LLaMA 8B in math—beating a model 8x its size.…

115 624 5K 495K 2K

Download Image

Yuchen Zeng @yzeng58

10 months ago

🎉 Milestone: Our LIFT paper has hit 100+ citations! We introduced a simple method to adapt LLMs to new domains, and researchers are now achieving success with it across predictive chemistry, metamaterial physics & more! Check our work at uw-madison-lee-lab.github.io/LanguageInterf…

1 14 99 21K 37

Kangwook Lee @Kangwook_Lee

11 months ago

😎TLDR😎 LLMs can simultaneously solve many in-context learning tasks! How? By giving the LLM a randomly shuffled examples from multiple tasks! This super fun project all started with the out-of-the-box thinking of @DimitrisPapail and great team effort led by @zheyangxiong!

Dimitris Papailiopoulos @DimitrisPapail

11 months ago

16 122 643 80K 575

Download Image

1 6 60 6K 10

Kangwook Lee @Kangwook_Lee

11 months ago

🚀 Excited to share our latest research on Looped Transformers for Length Generalization! TL;DR: We trained a Looped Transformer that dynamically adjusts the number of iterations based on input difficulty—and it achieves near-perfect length generalization on various tasks! 🧵👇