Roy Xie (@RoyXie

Roy Xie

100 posts

Roy Xie

@RoyXie_

PhD Student @DukeU @duke_nlp | Apple Scholar | @NSF GRFP | Research intern @Apple Foundation Models

Joined June 2021

Pinned
Roy Xie
@RoyXie_
May 13
VLMs can compress text by rendering it as images, but accuracy collapses once images shrink below a certain resolution. We introduce LensVLM: teach the model to scan compressed images, then selectively decompress what it needs. Paper: arxiv.org/abs/2605.07019
6.9K
Roy Xie
@RoyXie_
May 27, 2025
Can we train reasoning LLMs to generate answers as they think? Introducing 𝐈𝐧𝐭𝐞𝐫𝐥𝐞𝐚𝐯𝐞𝐝 𝐑𝐞𝐚𝐬𝐨𝐧𝐢𝐧𝐠! We train LLMs to alternate between thinking & answering 🚀 Reducing Time-to-First-Token (TTFT) by over 80% ⚡AND improving Pass@1 accuracy up to 19.3%!📈 🧵 1/n
17K
Roy Xie
@RoyXie_
Mar 30, 2025
Incredibly honored and grateful to receive the 2025 Apple Scholars in AI/ML PhD Fellowship! Huge thanks to @Apple for the recognition, and to all my amazing collaborators, mentors, and especially my advisor @bhuwandhingra for the invaluable support!
machinelearning.apple.com
Announcing the 2025 Apple Scholars in AIML
Apple is proud to announce the 2025 recipients of the Apple Scholars in AIML PhD fellowship.
13K
Roy Xie
@RoyXie_
Mar 18, 2025
Excited to share that I will be joining @Apple Seattle as a Research Intern, working on LLM reasoning/efficiency. Let me know if you would like to connect/meet up!
10K
Roy Xie
@RoyXie_
May 27, 2025
Thanks @_akhaliq for sharing our work!
AK
@_akhaliq
May 27, 2025
Apple presents Interleaved Reasoning for Large Language Models via Reinforcement Learning
14K
Roy Xie
@RoyXie_
Jan 28, 2023
What are the effective ways to align noisy data in under-served languages at the word level? 👉Check out our paper "Noisy Parallel Data Alignment", accepted at 𝗘𝗔𝗖𝗟 2023 (Findings)! arxiv.org/abs/2301.09685 (w/@anas_ant) 🧵1/8
7.3K
Roy Xie
@RoyXie_
Jun 25, 2024
🚨 Breaking: >90% AUC on the WikiMIA dataset for membership inference! Want to know if your data is in LLM's training set?🔍 Check out our latest work "ReCaLL: Membership Inference via Relative Conditional Log-Likelihoods" ✨ royxie.com/recall-project… 🧵1/6
5.2K
Roy Xie
@RoyXie_
Jun 16, 2024
Excited to be in CDMX for #NAACL2024🇲🇽! Presenting our work "Extracting Lexical Features from Dialects via Interpretable Dialect Classifiers" during Poster Session 3 on Monday (6/17) @ 4 PM. Swing by our poster & I'll share the best spots for tacos al pastor! 🌮
Roy Xie
@RoyXie_
Mar 4, 2024
✨ Can we use interpretability methods to extract linguistic features that characterize dialects❓ 🎉 New preprint: arxiv.org/abs/2402.17914 (@ruoyuxyz, @orevaahia, @tsvetshop, @anas_ant) 👉Code & Data: github.com/ruoyuxie/inter… 🧵(1/6)
2.4K
Roy Xie
@RoyXie_
Mar 4, 2024
✨ Can we use interpretability methods to extract linguistic features that characterize dialects❓ 🎉 New preprint: arxiv.org/abs/2402.17914 (@ruoyuxyz, @orevaahia, @tsvetshop, @anas_ant) 👉Code & Data: github.com/ruoyuxie/inter… 🧵(1/6)
9.3K
Roy Xie
@RoyXie_
Mar 13, 2024
📢 Excited to share that our paper has been accepted to #NAACL2024 main conference! 🌟Huge shoutout to amazing co-authors @orevaahia @tsvetshop @anas_ant! ⛱️ See you in Mexico City! 🇲🇽 #NLProc
Roy Xie
@RoyXie_
Mar 4, 2024
✨ Can we use interpretability methods to extract linguistic features that characterize dialects❓ 🎉 New preprint: arxiv.org/abs/2402.17914 (@ruoyuxyz, @orevaahia, @tsvetshop, @anas_ant) 👉Code & Data: github.com/ruoyuxie/inter… 🧵(1/6)
1.7K
Roy Xie
@RoyXie_
Nov 10, 2024
I'll be at #EMNLP2024🏖️ Happy to chat about anything, especially in LLM evaluation, adversarial robustness, and context compression. I will also be presenting two papers: 🧵
2.9K
Roy Xie
@RoyXie_
Apr 12, 2024
I'm excited to share that I've been awarded the NSF Graduate Research Fellowship! 🎉 I'm incredibly grateful to my advisors, mentors, and collaborators for their support. Looking forward to pursuing my research with this amazing opportunity #NSF #GRFP #NLProc
Duke CompSci
@dukecompsci
Apr 10, 2024
4 Duke CS students received NSF Grad Research Fellowships #nsfgrfp: Jon Donnelly-advisor C. Rudin; Jabari Kwesi-advisor P. Emami-Naeini, & Ruoyu (Roy) Xie, advisor B. Dhingra - all 3 will attend Duke; & Megan Richards, ECE-CS, advisors M. Sendak, R. Henao. research.gov/grfp
1.4K
Roy Xie
@RoyXie_
May 27, 2025
Replying to @RoyXie_
📢Huge thanks to my amazing co-authors: David Qiu, @DpacGopinath, @DongLin_AIML, @sun_yanchao, Chong Wang, @saloni_p, @bhuwandhingra
arxiv.org
Interleaved Reasoning for Large Language Models via Reinforcement Learning
Long chain-of-thought (CoT) significantly enhances the reasoning capabilities of large language models (LLMs). However, extensive reasoning traces lead to inefficiencies and increased...
363
Roy Xie
@RoyXie_
May 27, 2025
Replying to @RoyXie_
🌟𝘝𝘢𝘭𝘶𝘦 𝘪𝘯𝘴𝘪𝘨𝘩𝘵𝘴: 🔍 Generalization on OOD data & problems of varying difficulty. ⚙️ Effectiveness across PPO, GRPO, & REINFORCE++ algorithms. 📝 Analysis of reasoning dynamics. 🏆 Nuances of our conditional reward modeling. 🧵 6/n
414