Ryousuke Yamada

Can language models learn useful priors without ever seeing language? We pre-pre-train transformers on neural cellular automata — fully synthetic, zero language. This improves language modeling by up to 6%, speeds up convergence by 40%, and strengthens downstream reasoning. Surprisingly, it even beats pre-pre-training on natural text! Blog: hanseungwook.github.io/blog/nca-pre-p… (1/n)

@XiaoyangWu_ @_akhaliq Amazing work😀 Thanks for this great research

Thanks AK!

【4/7開催】ASPIRE Computer Vision Workshop @ Tokyo 日時：4月7日（火）10:30–18:00 会場：東京科学大学 大岡山キャンパス 蔵前会館 参加希望の方は以下より事前登録をお願いします。 research-p.com/event/2690 ※現地参加は定員制です。お早めにご登録ください。

@FumiyaUchiyama Amazing!! Congrats 🥳

Our paper "CLIP-like Model as a Foundational Density Ratio Estimator" has been accepted to #CVPR2026 (main)🎉 We reinterpreted CLIP/SigLIP as density ratio estimator and proposed applications for transfer learning / semantic diversity metric for an image! arxiv.org/abs/2506.22881

Our work on fine-grained Hand-Object Interaction understanding VideoQA benchmark (HanDyVQA) has been accepted to #CVPR2026 main track! We created 11K QA pairs and 4K segmentation masks that require holistic understanding of HOIs as dynamic process. Dataset already available!

Common knowledge in ML: more unique training data → better generalization. Turns out this doesn't hold for long-CoT SFT. Under a fixed update budget, repeating a small dataset multiple times beats training on more unique samples. And it's not even close.

@IdeKohsuke Congrats 👏

Two papers (1 main, 1 findings) have been accepted to CVPR 2026! 🎉 More details coming soon! #CVPR2026

@Masakichi333210 @CVPR Congrats 👏

Our paper "PowerCLIP: Powerset Alignment for Contrastive Pre-Training" has been accepted to @CVPR 2026! 🎉 See you in Denver!

@ulo_fd Wow, おめでとうございます👍

CVPR2本採択めでたい🎉

@shibatashi_ Congrats 🎉

Our paper has been accepted to CVPR 2026 main (this work was done during my visit to CMU)

@dahlian0 Congrats 🥳

My first author paper has been accepted to CVPR!🦋✨

貴重なニュルンベルク工科大学の体験記！ ドイツの生活がすごく伝わる記事でした、Airbnb含め、物件詐欺がよくあるので書いてあることは本当に参考になるし、気をつけるべき点など把握できそうです・・・！ x.com/FragileGoodwil…

今更ながら、海外研究留学 Advent Calendar 2025にお誘いいただき、寄稿しました!! adventar.org/calendars/12626 ドイツ（ニュルンベルク）での家探し＆一人暮らし立ち上げの話を書きました🇩🇪 現地での仮住まいなどリアル体験をまとめています note.com/ryousukeeee/n/…

生データを見ろおじさん｜八木 拓真 @t_Signull cvpaper.challenge アドベントカレンダー 11日目の記事になります。 #CVPAdv2025 note.com/takumayagi/n/n…

🚀 New arXiv preprint! PowerCLIP is the first method to align **powersets of image region subsets with textual phrase structures**, enabling fine-grained compositional and robust image-text understanding beyond simple global or token-to-patch alignment.

cvpaper.challenge アドカレ1日目で記事を執筆させていただきました！ note.com/ryousukeeee/n/…

@HoursWaking Thank you😀

@FragileGoodwill Congrats! Thanks for your contribution to our community.

I’m honored to be selected as an Outstanding Reviewer for #BMVC2025 😀

@ljihng9254221 Thank you so much!! No, but you can check the following link bmvc2025.bmva.org/people/reviewe…

@FragileGoodwill 🎉Congratulations! Did you get the notification email?

@shunk031 Dankeschön!!

@FragileGoodwill Sugoi! Congrats 🎉