Heuristic Solutions for Steepest Descent on the Stiefel Manifold

What would Muon look like if we constrained the weights to be semi-orthogonal?

July 18, 2025 · 28 min · Franz Louis Cesista

Training Transformers with Enforced Lipschitz Bounds

Neural networks are often highly sensitive to input and weight perturbations. This sensitivity has been linked to pathologies such as vulnerability to adversarial examples, divergent training, and overfitting. To combat these problems, past research has looked at building neural networks entirely from Lipschitz components. However, these techniques have not matured to the point where researchers have trained a modern architecture such as a transformer with a Lipschitz certificate enforced beyond initialization. To explore this gap, we begin by developing and benchmarking novel, computationally-efficient tools for maintaining norm-constrained weight matrices. Applying these tools, we are able to train transformer models with Lipschitz bounds enforced throughout training. We find that optimizer dynamics matter: switching from AdamW to Muon improves standard methods – weight decay and spectral normalization – allowing models to reach equal performance with a lower Lipschitz bound. Inspired by Muon’s update having a fixed spectral norm, we co-design a weight constraint method that improves the Lipschitz vs. performance tradeoff on MLPs and 2M parameter transformers. Our 2-Lipschitz transformer on Shakespeare text reaches validation accuracy 60%. Scaling to 145M parameters, our 10-Lipschitz transformer reaches 21% accuracy on internet text. However, to match the NanoGPT baseline validation accuracy of 39.4%, our Lipschitz upper bound increases to 10^264. Nonetheless, our Lipschitz transformers train without stability measures such as layer norm, QK norm, and logit tanh softcapping.

July 17, 2025 · 2 min · Laker Newhouse* and R. Preston Hess* and Franz Cesista* and Andrii Zahorodnii and Jeremy Bernstein and Phillip Isola

Sensitivity and Sharpness of n-Simplicial Attention

Towards a maximal update parameterization of n-simplicial attention

July 6, 2025 · 23 min · Franz Louis Cesista

Adam with Aggressive Gradient Clipping ≈ Smoothed SignSGD/NormSGD

Why does Adam with aggressive gradient value/norm clipping have sparse updates and do well with higher learning rates? Here we show that it is essentially equivalent to a smoothed version of SignSGD/NormSGD.

July 3, 2025 · 7 min · Franz Louis Cesista

Fast, Numerically Stable, and Auto-Differentiable Spectral Clipping via Newton-Schulz Iteration

A small step towards hardware-architecture-optimizer codesign in deep learning.

June 23, 2025 · 32 min · Franz Louis Cesista