🔥The 9th Round of Easy Loan, Earn $40 Reward is in progress❗️ ⏰ Promotion Period: January 15th - Feburary 15th, 2025 👉 Register now and check more details at gate.io/campaigns/358

At the very last day of this decade our paper w Omar, Seokho and Kyunghyun Cho on molecular geometry prediction with generative graph neural nets has been published to Scientific Reports. Check it out here nature.com/articles/s4159…. Happy New Year everyone!

The "Deep Learning Toolbox" has greatly expanded in the last decade thanks to our wonderful research community. Also, important progress has been made to make our community more inclusive and less toxic. Still, there's LOTS to do, and I plan to keep focusing on advancing both.

Some representative references on each group with more exciting approaches missed and more to come soon!

Our poster on latent-variable non-autoregressive machine translation will be presented at NLP 244 at American Hall 1, come over to have a talk #AAAI2020

“On the Discrepancy between Density Estimation and Sequence Generation” arxiv.org/pdf/2002.07233… Seq2seq models are optimized w.r.t log-likelihood. We investigate the correlation btw. LL and generation quality on machine translation. w/ Dustin Tran, Orhan Firat, Kyunghyun Cho. (1/n)

Animated attention by Kyunghyun Cho which took his good 3hrs @facebookai PyTorch #torched #deeplearning #nlp #attention #seq2seq #kyunghyuncho #nyu #cds

弊研の朱中元君（博士修了）が情報理工の研究科長賞を受賞しました。おめでとうございます！

これまでは東大の計算クラスタを利用したが、同じペースでABCIで実験をやる場合、月150～280万円かかるという事実がわかった。

空港は人がほぼ見当たらなくて、逆に怖いですね。

How Shall Tech Companies Navigate The Coronavirus Pandemic, Some of My Thoughts: nlper.com/2020/04/16/how…

Happy to announce our most recent paper: BLEU might be Guilty but References are not Innocent - arxiv.org/abs/2004.06063 - We show that it is possible to calculate reliable automatic scores (even with BLEU) for high quality MT output by using a novel reference generation method.

New work: "Iterative Refinement in the Continuous Space for Non-Autoregressive Neural Machine Translation" w/ Raphael Shu Kyunghyun Cho arxiv.org/pdf/2009.07177… We propose an inference procedure for non-autoregressive NMT that refines translations *purely in the continuous space.*

Summary: refining in the continuous space works for discrete output also! 😃 accepted to EMNLP 2020 (n/n)

Gmail down, can't work, Youtube down, can't play

Oh nice!

Pray that the auto driving car will be okay.

Happy to share our new #BLEURT-based reward optimization using a ranking-based contrastive margin loss for #NMT. We reconfirm the limitation of BLEU from our reward optimization results and human evaluation. Congrats to Raphael Shu, Kangmin Yoo. arxiv: arxiv.org/abs/2104.07541

ai.googleblog.com/2022/01/lamda-…

Just attended Multi-Agent Workshop at AAAI 2025 and it's hands down the best workshop I've been to at AAAI! It was buzzing and truly inspiring. Huge shoutout to the organizers (Weiyan Shi@ICLR and CHI Raphael Shu) for putting together such an unforgettable experience! #AAAI2025