Pith. sign in

REVIEW 11 cited by

A New Perspective on Shampoo's Preconditioner

Not yet reviewed by Pith; the record is open.

This paper has not been read by Pith yet. Machine review is queued; the pith claim, tier, and objections will appear here once it completes.

SPECIMEN: schema-true, not a live event

T0 review · schema-true

One-sentence machine reading of the paper's core claim.

pith:XXXXXXXX · record.json · timestamp

arxiv 2406.17748 v1 pith:B3XVLFOH submitted 2024-06-25 cs.LG math.OCstat.ML

A New Perspective on Shampoo's Preconditioner

classification cs.LG math.OCstat.ML
keywords approximationshampookroneckerproducthessianoptimalpreconditioneralgorithm
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved
0 comments
Share X Bluesky LinkedIn Reddit HN
abstract

Shampoo, a second-order optimization algorithm which uses a Kronecker product preconditioner, has recently garnered increasing attention from the machine learning community. The preconditioner used by Shampoo can be viewed either as an approximation of the Gauss--Newton component of the Hessian or the covariance matrix of the gradients maintained by Adagrad. We provide an explicit and novel connection between the $\textit{optimal}$ Kronecker product approximation of these matrices and the approximation made by Shampoo. Our connection highlights a subtle but common misconception about Shampoo's approximation. In particular, the $\textit{square}$ of the approximation used by the Shampoo optimizer is equivalent to a single step of the power iteration algorithm for computing the aforementioned optimal Kronecker product approximation. Across a variety of datasets and architectures we empirically demonstrate that this is close to the optimal Kronecker product approximation. Additionally, for the Hessian approximation viewpoint, we empirically study the impact of various practical tricks to make Shampoo more computationally efficient (such as using the batch gradient and the empirical Fisher) on the quality of Hessian approximation.

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Forward citations

Cited by 11 Pith papers

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score.

  1. Optimization Geometrodynamics: Variational Reduction and Interaction Curvature

    math.OC 2026-07 conditional novelty 7.0

    Determinant-one SPD action fibers form a global analytic totally geodesic bundle whose AIRM nearest controller is computable by a residual-gradient iteration with posterior certificates and recovers inverse-Hessian sh...

  2. Why Muon Outperforms Adam: A Curvature Perspective

    cs.LG 2026-06 conditional novelty 7.0

    Muon outperforms Adam by reducing curvature penalty via lower Normalized Directional Sharpness, as shown via Taylor approximation on LLM training and proven on stylized quadratic problems with heterogeneous curvature.

  3. Old Optimizer, New Norm: An Anthology

    cs.LG 2024-09 unverdicted novelty 7.0

    Optimizers like Adam reduce to steepest descent under particular norms, opening a design space of norm assignments tailored to layer roles.

  4. SOAP, Muon, and Beyond: Pushing LLM Pretraining Scales

    cs.LG 2026-07 conditional novelty 6.0

    SOAP and Muon, stabilized by per-step QR eigenbasis updates and KL-Shampoo covariance accumulation, beat AdamW on large-batch LLM pretraining up to 100M-token batches.

  5. Causal Optimizer Interaction Calculus: Hidden Geometric Relaxation and Identifiable Interventions

    cs.LG 2026-07 conditional novelty 6.0

    Under fixed innovation coupling, finite-horizon optimizers admit minimal pathwise realizations and incidence-identifiable Möbius effects, with a five-term readout transfer from hidden relaxation and a closed reduced-v...

  6. Optimization Geometrodynamics: Variational Reduction and Interaction Curvature

    math.OC 2026-07 accept novelty 6.0

    Dynamic geometric complexity of reducing condition number under full SPD metric control equals the affine-invariant distance from the relative log-spectrum to a low-width set.

  7. Pro-KLShampoo: Projected KL-Shampoo with Whitening Recovered by Orthogonalization

    cs.LG 2026-05 unverdicted novelty 6.0

    Pro-KLShampoo projects KL-Shampoo preconditioners to a spike-and-flat parametric form on an r-dimensional subspace and recovers the full algebraic preconditioner via orthogonalization, outperforming KL-Shampoo on GPT-...

  8. Decoupling Variance and Scale-Invariant Updates in Adaptive Gradient Descent for Unified Vector and Matrix Optimization

    cs.LG 2026-02 conditional novelty 6.0

    DeVA_S8 reweights Muon's matrix-sign update in the matrix's eigenbasis with a singular-value signal-to-noise ratio, reaching target LLM validation perplexity with ~6.6% fewer tokens than Muon.

  9. Harnessing Optimization Dynamics for Curvature-Informed Model Merging

    cs.LG 2025-09 conditional novelty 6.0

    Optimization Trajectory Aware merging uses Adam second moments as a curvature proxy, first pruning task-vector edits with Fast Fisher Grafting, then reweighting survivors with a compressed curvature preconditioner.

  10. Causal Optimizer Interaction Calculus: Hidden Geometric Relaxation and Identifiable Interventions

    cs.LG 2026-07 conditional novelty 5.0

    A modular calculus decomposes optimizer updates into geometric preconditioning plus structured nongeometric mechanisms, with a direction-expressivity theorem showing full SPD geometry captures exactly strict descent d...

  11. On the Convergence Analysis of Muon

    stat.ML 2025-05 unverdicted novelty 5.0

    Convergence analysis shows Muon outperforms gradient descent by exploiting low-rank structure in neural network Hessians.