noukou dashi soup

Broth vs stock

[noukou dashi soup @noukoudashisoup]

"Because you are paid only for doing research, the pressure is terrible." https://t.co/B1BRrXLQBx

[Jon Barron @jon_barron]

Good news everyone, kernel methods are AGI, pack it up

[Machine Learning (ML) Papers @Memoirs]

Thinned Mean Field Langevin Dynamics

Zonghao Chen, Heishiro Kanagawa, François-Xavier Briol, Chris J. Oates, Lester Mackey
https://t.co/nFRkJK6giU [𝚌𝚜.𝙻𝙶]

[Hudson @Hudson19990518]

🚀 New paper at ICML 2026!

For Wasserstein gradient flows with N particles, the cost is N^2 due to interactions.
We reduce the cost from N^2 to N^{1.5}, without sacrificing the quality simulated until convergence.

The key idea is kernel thinning! ✨
📄 https://t.co/zXfflTfxLt

[noukou dashi soup @noukoudashisoup]

Thought it was Keisuke Matsushima

[Umami Lab @umamilabldn]

今月Camdenにオープンした
北海道出身オーナーによる札幌ラーメン店「Umami Lab」公式アカウントです。

味噌、塩、醤油、混ぜそば——
ロンドンで本気の札幌ラーメンを作ってます。どうぞよろしくお願いします🍜

[noukou dashi soup @noukoudashisoup]

https://t.co/A0oM1totRr

[Kenji Fukumizu @KFukumizu]

文部科学大臣表彰・科学技術賞（研究部門）をいただくことができました。共同研究者やご支援いただく方々のご協力があればこその受賞と思います。今後も一層精進したいと思います

[Lester Mackey @LesterMackey]

Qiang Liu, Chris Oates, and I are writing a monograph on Probabilistic Inference and Learning with Stein’s Method, and we’d love to get your feedback on the first draft

[noukou dashi soup @noukoudashisoup]

Part of the work has also been used for a generalisation of the KSD, which quantifies the size of "distributional gradient": https://t.co/ztO3QRG5xf

[noukou dashi soup @noukoudashisoup]

Our paper on KSD-Wasserstein convergence was published in Annals of Applied Probability.
Grateful to my co-authors (Alessandro Barp, @ArthurGretton, @LesterMackey ) and editors/reviewers for the helpful feedback.

https://t.co/kYxpt0PuMP

[noukou dashi soup @noukoudashisoup]

@etha1_1_ お返事ありがとうございます（今気づきました）
PhDのところと、ボスのところ、会社がありそうなところが自分と完全一致しててわろてしまいました😂

ちなみにイギリスから日系の研究所で働くというルートもあります！

[noukou dashi soup @noukoudashisoup]

@etha1_1_ フォロー外から失礼します、何系のPhDの方の話だったかわかりますか？

[noukou dashi soup @noukoudashisoup]

I can fit all my personal belongings into two suitcases. This is known as the universal property in the literature.

[noukou dashi soup @noukoudashisoup]

Newcastle is great. You should come.

[noukou dashi soup @noukoudashisoup]

don't blame yourself, blame the job market

[noukou dashi soup @noukoudashisoup]

One man’s meat is another man’s poison.

[noukou dashi soup @noukoudashisoup]

@erfunmirzaei @PontilGroup nice one

[noukou dashi soup @noukoudashisoup]

@xidulu @FlatironInst Was not aware! Thank you!

[noukou dashi soup @noukoudashisoup]

This is joint work between two teams:
• Clémentine Chazal and @Korba_Anna (@ENSAEparis)
• Heishiro Kanagawa, @zheyangshen, and Chris Oates (@uniofnewcastle)
Part of this work was carried out during Clémentine’s (the student author!) visit to Newcastle as a visiting fellow.

[noukou dashi soup @noukoudashisoup]

In optimisation, gradients guide convergence.
But when the objective is over probability distributions, what’s the analogue?
This paper introduces a way to evaluate sample approximations to optima for entropy-regularised objectives:

https://t.co/ztO3QRG5xf

[noukou dashi soup @noukoudashisoup]

At the end of the day, we obtain and use sample approximations, but their KL is undefined (objective cannot be computed).
This work thus presents a computable alternative termed 'gradient discrepancies', which turns out to generalise the kernel Stein discrepancy!