I will attend #CVPR2024 next week to present our paper "Adaptive Random Feature Regularization on Fine-tuning Deep Neural Networks" at the main conference (Poster Session 5)! In this work, we propose an effective and efficient regularization method for fine-tuning.

2022年10月に発足したNTT基礎数学研究センタの構成員による特集。どれも面白い。 journal.ntt.co.jp/article/27040

now arrived at Vienna because I missed my connecting flight this morning😂

#NeurIPS2023 論文"Compositional Abilities Emerge Multiplicatively: Exploring Diffusion Models on a Synthetic Task"のビデオがYouTube で公開されました。ぜひ見てください！ Diffusion Model - Maya Okawa youtu.be/qI3NLutMlx8?si… via YouTube

Also, this work will be presented at the MINT workshop at #NeurIPS2024 tomorrow! Please come to my poster at West Meeting Room 121, 122 (1:00-2:00 PM)😎

Tomorrow, I will present this work about the Strong Lottery Ticket Hypothesis at the compression workshop (West Meeting Room 211-214) of #NeurIPS2024 😁 Let's talk with me about SLTH💪 arxiv.org/abs/2402.14029

🚀New paper on Theory of Mind accepted AAAI🚀 We present 🍅ToMATO, a benchmark for comprehensively evaluating Theory of Mind in LLMs in a more practical setting. 🔗 Paper: arxiv.org/abs/2501.08838 🔗 Data&Code: github.com/nttmdlab-nlp/T… #AAAI25 #AAAI2025

Finally, our paper about Strong Lottery Tickets within Frozen Networks has been accepted & published at #TMLR! I would like to thank the equally contributed authors, Chijiwa-san (Daiki Chijiwa) and López-san, and all the co-authors for their cooperation! openreview.net/forum?id=xpnPY…

And finally one more open access one: "Transfer learning with pre-trained conditional generative models" by Shin’ya Yamaguchi, Sekitoshi Kanai, Atsutoshi Kumagai, Daiki Chijiwa & Hisashi Kashima (link.springer.com/article/10.100…) #OA

AI is reshaping society and ourselves. We're at the wheel, steering this powerful force. But what future do we envision for a human-AI society, and how can we get there safely? We’ve launched "Physics of AI Group" to explore the Philosophy and Physics of Intelligence.

I'm happy to announce that our paper, "Test-time Adaptation for Regression by Subspace Alignment," will be presented at #ICLR2025 on 4/25! We demonstrate that existing TTA methods for classification struggle with regression and propose an effective method using feature alignment.

We have released the preprint and code for our paper, which has been accepted to #CVPR2025! We propose a simple post-pre-training method called CLIP-Refine to mitigate the modality gap in CLIP. Preprint: arxiv.org/abs/2504.12717 Code: github.com/yshinya6/clip-…

🎉 Excited to announce our ICML 2025 paper “Portable Reward Tuning: Towards Reusable Fine‑Tuning across Different Pretrained Models,” co‑first‑authored with Daiki Chijiwa 🤝(equal contribution)! #ICML2025 Preprint 👉 arxiv.org/abs/2502.12776

🚀 Happy to share our #CVPR2025 paper, which proposes a post-pre-training method for CLIP to mitigate modality gaps and improve zero-shot performance with just 5 minutes of additional training! We are looking forward to discussing with you at our poster session! #CVPR2025

🚀次回の #NLPコロキウム のお知らせ 講演者: 長谷川拓さん Taku Hasegawa Daiki Chijiwa (NTT) 日時: 7/9 (水) 12:00–13:00 JST 世界の変化に応じたモデル更新には高いコストが伴います🌏そこで基盤モデルに柔軟に組み合わせて使えるポータブルな報酬モデルを学習するお話です。nlp-colloquium-jp.github.io/schedule/2025-…

セキュリティに配慮した #LLM の応答精度向上技術を確立しました 定型的な自動応答においてLLMの応答からの学習用データの漏洩リスクを抑えつつ応答精度向上に活用が期待されます #NTTRD

✨世界初✨ #生成AI のカスタマイズコストを抜本的に低減し、低コスト運用を持続可能にする「ポータブルチューニング」技術を確立しました！ 本成果は、7/13～19に開催される機械学習分野における最難関国際会議 #ICML2025 にて発表されます #NTTRD

先日、NLPコロキウムでお話しさせていただいた動画がアップロードされました。 ご興味がある方はどうぞご覧ください。 内容はICML2025 に採択された Portable Reward Tuning のお話です。