Byeongho Heo

Research scientist, NAVER AI Lab

[Sangdoo Yun @oodgnas]

On-policy distillation (OPD) makes the student match the teacher's output distribution, including the style and preferences it had *before* reasoning tuning.

We distill only what reasoning tuning added.

📄 On-Policy Delta Distillation (OPD²)
https://t.co/mzRDnSd17q
(1/n)

[Liner @search_liner]

Liner is partnering with @spoticlr at #ICLR2026 — supporting Best Paper and Travel Awards for LLM research.

And to celebrate, we're giving away:
✈️ Round-trip flights + hotel to #ICML2026 in Seoul
🎁 $300 Liner Credits

Follow @search_liner + repost to enter by 4/27.

Liner is built for research workflows. Find papers, verify sources, and write with citations in one place.

See you in 🇧🇷 and 🇰🇷!

@iclr_conf @icmlconf

[Byeongho Heo @bhheo90]

TL;DR by first author @JunhaSong:
https://t.co/N8P2ImvHYq

[JunhaSong @junha_song]

🌎𝐖𝐨𝐫𝐥𝐝: "Hail the LLMs" 🥶 𝐕𝐢𝐬𝐢𝐨𝐧 𝐞𝐧𝐜𝐨𝐝𝐞𝐫: "Hello? Still here" ⚡𝐑𝐋: "Hold my (beer) gradients"
Proud to share our new paper — "𝐑𝐋 𝐦𝐚𝐤𝐞𝐬 𝐌𝐋𝐋𝐌𝐬 𝐬𝐞𝐞 𝐛𝐞𝐭𝐭𝐞𝐫 𝐭𝐡𝐚𝐧 𝐒𝐅𝐓."
🔗 Project-page: https://t.co/6P65bImapb

1/4

[Byeongho Heo @bhheo90]

We just released our new paper on how RL impacts MLLM training, especially focused on the vision encoder.

Paper: https://t.co/VJnZFZruVD
Project page: https://t.co/wWBIDOIaME

[JunhaSong @junha_song]

🌎𝐖𝐨𝐫𝐥𝐝: "Hail the LLMs" 🥶 𝐕𝐢𝐬𝐢𝐨𝐧 𝐞𝐧𝐜𝐨𝐝𝐞𝐫: "Hello? Still here" ⚡𝐑𝐋: "Hold my (beer) gradients"
Proud to share our new paper — "𝐑𝐋 𝐦𝐚𝐤𝐞𝐬 𝐌𝐋𝐋𝐌𝐬 𝐬𝐞𝐞 𝐛𝐞𝐭𝐭𝐞𝐫 𝐭𝐡𝐚𝐧 𝐒𝐅𝐓."
🔗 Project-page: https://t.co/6P65bImapb

1/4

[국가AI전략위 부위원장 하정우 @JungWooHa2]

So happy to annouce that #NAVER Cloud released our own Korean-aware multimodal multilingual foundation model, #HyperCLOVA_SEED via hugging face with #commercial license. The models include 3B, 1.5B, and 0.5B parameter models.

You can find them here:
https://t.co/eaSdZQS4pW

[Byeongho Heo @bhheo90]

Our paper "Masking meets Supervision: A Strong Learning Alliance" has been accepted to #CVPR2025

We propose MaskSub, a novel self-distillation based supervised learning framework with strong masking augmentation.

Paper: https://t.co/sNkdwi1Bc8
Code: https://t.co/9AkgXtYvJz

[Young-Ho Kim @YounghoHCI]

(🧵1/7) 🚀Excited to share our #CHI2025 preprint on ExploreSelf, an #LLM-guided tool supporting user-driven reflection and exploration of personal challenges, an internship project of @NAVER_AI_Lab led by @_inhwa_song!
#HCI #AI #NLProc #MentalWellness #DigitalWellbeing

[Young-Ho Kim @YounghoHCI]

🧵(1/7)🚀 Excited to share our #CHI2025 preprint on ELMI, an LLM-infused interactive & intelligent sign language translation of lyrics for song-signing!

Work led by @suhyeon_yoo as a research intern at @NAVER_AI_Lab. #AI #Accessibility #HCI #NLProc

[Young-Ho Kim @YounghoHCI]

🚀Excited to share our #CHI2025 paper on AACessTalk, an #AI-driven tablet app that fosters communication between a minimally-verbal autistic (MVA) child and parent, which was an internship project of @NAVER_AI_LAB led by @manycotton94! 🧵(1/6) #hci #nlproc #autism @acm_chi

[Byeongho Heo @bhheo90]

@bronzeagepapi We just covered image recognition models. There is no generative model or encoder-decoder result.

[Byeongho Heo @bhheo90]

I'm happy to share that my paper, "Rotary Position Embedding for Vision Transformer," has been accepted to #ECCV2024

We demonstrate the benefits of RoPE when it is applied to ViT.

Paper: https://t.co/Kov2r2I70S
Code: https://t.co/Vxu4IuBkCe

[Seong Joon Oh @coallaoh]

Exploring the potential of Representation Learning in transfer tasks with few shot samples. Do "Probabilistic Representations" aid downstream uncertainty estimation? @mkirchhof_ delivers the answer with a new benchmark: Uncertainty-aware Representation Learning (URL). #AI

[Sangdoo Yun @oodgnas]

Excited to introduce our new paper, 'Augmenting Sub-model to Improve Main Model' (AugSub)! 🚀

Paper: https://t.co/4SngnhSCXJ
Code: https://t.co/PhK7neun1z

We've been tackling an ongoing problem -- conducting strong regularization *without* harming loss convergence. 🔍 [1/3]

[Byeongho Heo @bhheo90]

시작 !!