Wu Lin

Postdoctoral fellow at @VectorInst. ML PhD at UBC. Mathematical and computational structures for ML. Geometric and algebraic methods.

[Wu Lin @LinYorker]

#ICML2024
Can We Remove the Square-Root in Adaptive Methods?
https://t.co/hD604GmB0N
Root-free (RF) methods are better on CNNs and competitive on Transformers compared to root-based methods (AdamW)

Removing the root makes matrix methods faster: Root-free Shampoo in BFloat16 /1

[Wu Lin @LinYorker]

@MarktHart125849 I mean this repo. https://t.co/2XWz4yYJgC

[Wu Lin @LinYorker]

On one hand, it is essential to tune baseline methods well on a model. On the other hand, it may be better to avoid using a model/architecture that has been modified and optimized for a single method for 1.5 years.

[Konstantin Mishchenko @konstmish]

I just submitted a PR to modded-nanogpt with better hyperparams. With them, Muon can reach the target loss after 3250 steps instead of 3325.
Always tune your baseline well when doing research. Weak baselines can make any idea look promising

[Wu Lin @LinYorker]

Some initial steps to make Shampoo and SOAP faster https://t.co/wRGcI92w3W We are working on further improvements.

[Wu Lin @LinYorker]

We will make Shampoo/SOAP, including KL-Shampoo/KL-SOAP, faster. Our goal is to match Muon's runtime while maintaining Shampoo/SOAP's strong per-step performance. Stay tuned for new updates.

[Lucas Nestler @Clashluke]

KL Shampoo and KL SOAP outperform their non-KL counterparts by learning the preconditioners compositionally, so that each stage corrects what remains after the last.

Available in HeavyBall 3.1.1, with major PSGD stability backports.

[Wu Lin @LinYorker]

@wen_kaiyue You may want to have a look at this paper for further improvement https://t.co/wRGcI92w3W

[Wu Lin @LinYorker]

@_arohan_ Muon

[Wu Lin @LinYorker]

also, short sided KL-Shampoo = short-sided Shampoo^2 = Muon

[Wu Lin @LinYorker]

@weijie444 Looks like a KFAC-based method with modern clipping? G(ZZ^T)^{-1} is known as the FOOF update https://t.co/Bgh2m79ZTK while msgn() can be interpreted as "generalized (preconditioned) gradient norm clipping" https://t.co/DPyGmHecz1 .

[Wu Lin @LinYorker]

@weijie444 Looks like a KFAC-based method with modern clipping? G(ZZ^T)^{-1} is known as the FOOF update https://t.co/Bgh2m79ZTK while msgn() can be interpreted as "generalized (preconditioned) gradient norm clipping" https://t.co/DPyGmHecz1 .

[Weijie Su @weijie444]

We released "The Newton--Muon Optimizer" . We show that Muon is secretly an implicit Newton method, and use this insight to build a better one. 1/n

Paper: https://t.co/Ua54426bWB

[Wu Lin @LinYorker]

@MarkSchmidtUBC @EmtiyazKhan This work is a joint effort with Scott C. Lowe, @f_dangel, @runame_, Zikun Xu, and @RogerGrosse.
Stay tuned for more updates coming soon.

[Wu Lin @LinYorker]

Within an information-geometric framework, we reconnect Shampoo/SOAP with both classical quasi-Newton ideas and Gaussian whitening, and develop practical methods that naturally handle tensor-valued weights in language model pre-training. https://t.co/PJ4AVxPgRC opt-ml workshop

[Wu Lin @LinYorker]

This work builds on my ICML 2019 paper (with @MarkSchmidtUBC and @EmtiyazKhan), extending a variational Bayes-based geometric framework to modern NN optimization. It can be used to design methods for Bayesian inference, numerical optimization, and gradient-free optimization.

[Runa Eschenhagen @runame_]

1/9 In practice, the Shampoo optimizer crucially relies on several heuristics.

In our NeurIPS 2025 spotlight paper, we investigate the role of learning rate grafting and infrequent preconditioner updates in Shampoo by decomposing its preconditioner.

https://t.co/TfI1gwMrFs