🔥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

Amazing lecture notes!

Beside being useful to prove Weierstrass polynomial approximation theorem quantitatively, Bernstein polynomial approximation enjoys many desirable properties. en.wikipedia.org/wiki/Bernstein…

Very beautiful mathematical objects!

Exciting start of the year for theory of #DeepLearning! SGD on neural nets can: 1) simulate any other learning alg w. some poly-time init [Abbe & Sandon arxiv.org/abs/2001.02992] 2) learn efficiently hierarchical concept classes [Zeyuan Allen-Zhu, Sc.D. & Y. Li arxiv.org/abs/2001.04413]

Dear Reviewers: Don't reject because we made it look easy. Making it look easy is hard work. Don't reject because our solution is simple. Simple solutions often have the most impact. Don't reject because you can imagine additional experiments. There's a page limit.

Our requests have been listened! New dates for NeurIPS Conference Abstract submission deadline: May 27 Paper submission deadline: June 3 Author notification date: Sept 25

Big News! I am very excited to announce that Ashok Cutkosky and I will give a tutorial at ICML Conference 2020 on Parameter-free Online Optimization! We will teach you how to get rid of learning rates, for real :) Please spread the word around! #icml2020 #tutorial #MachineLearning

I take a look at the table of contents and have finally realized that there are so many celebrated results obtained by Dr. Lan and his co-authors. Great book!

When I was a naive starting grad student, I thought the goal was to publish as many papers as possible. Wrong for many reasons, but one: if people see a *bad* paper you wrote, they will remember it. And they'll be prejudiced against your future work. Write papers you're proud of.

Why Does Deep Learning Perform Deep Learning - MSR AI Seminar 08/11/2020 youtu.be/sd2o1PbqixI

Are you hesitant to review for #NeurIPS2022 because you cannot commit to reviewing 5 papers? Worry no more! In our blog, we explain how you can request a reduced workload. Thank you for helping the community. blog.neurips.cc/2022/04/01/how…

ICML meta reviews are just arbitrary! In rebuttal we did exactly what reviewers asked, but AC rejected saying: "it is not certain whether this new experiment performed in a short period of time was done accurately." If can reject by simply not trusting rebuttal, why having it?

Identify and remove the bad ACs! There is no shortage of qualified people who can AC! Let the authors and reviewers identify and report these bad ACs, and don’t invite them back!

As promised, here the paper in which, among the other things, we fix an error in a previously published lower bound that contained a log of a gradient. Accepted at #NeurIPS2022, congrats to my amazing co-authors! arxiv.org/pdf/2208.11195…

I have a poster today at #NeurIPS22 about optimization of Transformers. Hall J #413 Joint work with Michael Crawshaw, Mingrui Liu, Wei Zhang, and zhenxun zhuang. This was zhenxun zhuang's last work before his graduation this summer. A small 🧵 about it.

🚨New Blog Post🚨 Adapting to Smoothness with Normalized Gradients A "new" result:normalized gradients give adaptation to local Hölder smoothness,for free! The trick is general:any first-order optimization algorithm can be used, even parameter-free ones parameterfree.com/2023/06/19/ada…

I highly recommend Francesco Orabona! I learned a lot during my postdoc in his group—not only about research, but also about navigating the academic job market, writing clear and impactful papers, and mentoring students. Feel free to contact me: I'm happy to share my experience!