arXiv subjects

Maxim Rakhuba

Publications and source records attributed to Maxim Rakhuba.

At least 19 recordsLinked to original sources

arXiv2026

Low-rank approximation of Moment Tensor Potential enables reducing training set size without loss of accuracy

In this study, we implement a low-rank approximation of Moment Tensor Potential (MTP) based on the tensor train (TT) decomposition. The implemented tensor-factorized MTP (TFMTP) and the original MTP model are actively trained via a MaxVol-based algorithm during molecular dynamics simulations of a four-component molten salt mixture, LiF-NaF-KF (FLiNaK), and geometry optimizations of a five-component equiatomic MoNbTaWV random alloy. We demonstrate that under a 1.5-fold compression, TFMTP requires two times fewer configurations for fitting than the original MTP model, while maintaining an indistinguishable level of accuracy. These actively trained MTP and TFMTP models are further used to evaluate the density and viscosity of FLiNaK at temperatures ranging from 600 to 1200 K, as well as the elastic constants and bulk modulus of the MoNbTaWV alloy at zero temperature. For both atomic systems, the differences in physical properties predicted by MTP and TFMTP are negligible.

physics.chem-ph↗

arXiv2026

Muon on the Stiefel Manifold Admits an Exact Closed-Form Update

We study Muon, a recently proposed matrix-aware optimization method, in the context of the Stiefel manifold. This manifold consists of matrices with orthonormal columns and is ubiquitous in machine learning and scientific computing. Existing extensions of Muon to this manifold rely on heuristic, approximate, or iterative updates with varying computational efficiency. We show that the corresponding Stiefel Muon update admits an exact closed-form solution and use this result to develop Skewon, a practical algorithm for orthogonality-constrained optimization with an efficient implementation. We further establish first-order convergence guarantees for Skewon in the smooth non-convex setting.

math.OC↗

arXiv2026

Tensorion: A Tensor-Aware Generalization of the Muon Optimizer

Common first-order optimizers, such as Adam, implicitly treat each parameter block as an unstructured vector, which disregards the multilinear weight structure present in many modern machine learning models. Recent work has shown that exploiting matrix structure can improve optimization dynamics. A notable example is Muon, which performs steepest descent under the spectral norm constraint. We take the next step and introduce Tensorion, a tensor-aware optimizer that extends Muon's constrained optimization perspective from matrices to higher-order tensors. Tensorion is built around a linear minimization oracle (LMO) over a tensor norm ball. The norm is carefully chosen to balance two objectives: tightly bounding the tensor spectral norm, while still keeping the LMO tractable. This LMO becomes computable because it reduces to operations on adaptively selected unfolding matrices. Notably, when restricted to order-2 tensors (i.e., matrices), Tensorion recovers Muon exactly. Experiments on tensor-based computer vision problems suggest that Tensorion can offer improved convergence behavior and more stable gradient updates compared with Adam-based and existing tensor-aware baselines in the evaluated settings.

cs.LG↗

arXiv2026

OrthoFuse: Training-free Riemannian Fusion of Orthogonal Style-Concept Adapters for Diffusion Models

In a rapidly growing field of model training there is a constant practical interest in parameter-efficient fine-tuning and various techniques that use a small amount of training data to adapt the model to a narrow task. However, there is an open question: how to combine several adapters tuned for different tasks into one which is able to yield adequate results on both tasks? Specifically, merging subject and style adapters for generative models remains unresolved. In this paper we seek to show that in the case of orthogonal fine-tuning (OFT), we can use structured orthogonal parametrization and its geometric properties to get the formulas for training-free adapter merging. In particular, we derive the structure of the manifold formed by the recently proposed Group-and-Shuffle ($\mathcal{GS}$) orthogonal matrices, and obtain efficient formulas for the geodesics approximation between two points. Additionally, we propose a $\text{spectra restoration}$ transform that restores spectral properties of the merged adapter for higher-quality fusion. We conduct experiments in subject-driven generation tasks showing that our technique to merge two $\mathcal{GS}$ orthogonal matrices is capable of uniting concept and style features of different adapters. To the best of our knowledge, this is the first training-free method for merging multiplicative orthogonal adapters. Code is available via the $\href{https://github.com/ControlGenAI/OrthoFuse}{link}$.

cs.CV↗

arXiv2025

DyKAF: Dynamical Kronecker Approximation of the Fisher Information Matrix for Gradient Preconditioning

Recently, optimizers that explicitly treat weights as matrices, rather than flattened vectors, have demonstrated their effectiveness. This perspective naturally leads to structured approximations of the Fisher matrix as preconditioners, where the matrix view induces a Kronecker-factorized form that enables memory-efficient representation. However, constructing such approximations both efficiently and accurately remains an open challenge, since obtaining the optimal factorization is resource-intensive and practical methods therefore rely on heuristic design choices. In this work, we introduce a novel approach that leverages projector-splitting integrators to construct effective preconditioners. Our optimizer, DyKAF (Dynamical Kronecker Approximation of the Fisher Matrix), consistently improves the Fisher matrix approximation quality. Experiments on large language model pre-training and fine-tuning demonstrate that DyKAF outperforms existing optimizers across a range of evaluation metrics.

cs.LG↗

arXiv2025

Dimension-free Bounds for Covariance Estimation with Tensor-Train Structure

We consider a problem of covariance estimation from a sample of i.i.d. high-dimensional random vectors. To avoid the curse of dimensionality, we impose an additional assumption on the structure of the covariance matrix $\Sigma$. To be more precise, we study the case when $\Sigma$ can be approximated by a sum of double Kronecker products of smaller matrices in a tensor train (TT) format. Our setup naturally extends widely known Kronecker sum and CANDECOMP/PARAFAC models but admits richer interaction across modes. We suggest an iterative polynomial time algorithm based on TT-SVD and higher-order orthogonal iteration (HOOI) adapted to Tucker-2 hybrid structure. We derive non-asymptotic dimension-free bounds on the accuracy of covariance estimation taking into account hidden Kronecker product and tensor train structures. The efficiency of our approach is illustrated with numerical experiments.

math.ST↗

arXiv2025

Low-rank matrix and tensor approximations for compression of machine-learning interatomic potentials

Machine-learning interatomic potentials (MLIPs) have become a mainstay in computationally-guided materials science, surpassing traditional force fields due to their flexible functional form and superior accuracy in reproducing physical properties of materials. This flexibility is achieved through mathematically-rigorous basis sets that describe interatomic interactions within a local atomic environment. The number of parameters in these basis sets influences both the size of the training dataset required and the computational speed of the MLIP. Consequently, compressing MLIPs by reducing the number of parameters is a promising route to more efficient simulations. In this work, we use low-rank matrix and tensor factorizations under fixed-rank constraints to achieve this compression. In addition, we demonstrate that an algorithm with automatic rank augmentation helps to find a deeper local minimum of the fitted potential. The methodology is mainly verified using the Moment Tensor Potential (MTP) model and benchmarked on multi-component systems: a Mo-Nb-Ta-W medium-entropy alloy, molten LiF-NaF-KF, and a glycine molecular crystal. The proposed approach achieves up to 50 % compression without any loss of MTP accuracy. We also demonstrate that the developed methodology is universal and can be applied to compress other MLIPs on the example of Atomic Cluster Expansion (ACE).

physics.chem-ph↗

arXiv2025

Ultra Fast Warm Start Solution for Graph Recommendations

In this work, we present a fast and effective Linear approach for updating recommendations in a scalable graph-based recommender system UltraGCN. Solving this task is extremely important to maintain the relevance of the recommendations under the conditions of a large amount of new data and changing user preferences. To address this issue, we adapt the simple yet effective low-rank approximation approach to the graph-based model. Our method delivers instantaneous recommendations that are up to 30 times faster than conventional methods, with gains in recommendation quality, and demonstrates high scalability even on the large catalogue datasets.

cs.IR↗

arXiv2025

Optimization on the Extended Tensor-Train Manifold with Shared Factors

This paper studies tensors that admit decomposition in the Extended Tensor Train (ETT) format, with a key focus on the case where some decomposition factors are constrained to be equal. This factor sharing introduces additional challenges, as it breaks the multilinear structure of the decomposition. Nevertheless, we show that Riemannian optimization methods can naturally handle such constraints and prove that the underlying manifold is indeed smooth. We develop efficient algorithms for key Riemannian optimization components, including a retraction operation based on quasi-optimal approximation in the new format, as well as tangent space projection using automatic differentiation. Finally, we demonstrate the practical effectiveness of our approach through tensor approximation tasks and multidimensional eigenvalue problem.

math.NA↗

arXiv2025

Leveraging Geometric Insights in Hyperbolic Triplet Loss for Improved Recommendations

Recent studies have demonstrated the potential of hyperbolic geometry for capturing complex patterns from interaction data in recommender systems. In this work, we introduce a novel hyperbolic recommendation model that uses geometrical insights to improve representation learning and increase computational stability at the same time. We reformulate the notion of hyperbolic distances to unlock additional representation capacity over conventional Euclidean space and learn more expressive user and item representations. To better capture user-items interactions, we construct a triplet loss that models ternary relations between users and their corresponding preferred and nonpreferred choices through a mix of pairwise interaction terms driven by the geometry of data. Our hyperbolic approach not only outperforms existing Euclidean and hyperbolic models but also reduces popularity bias, leading to more diverse and personalized recommendations.

cs.IR↗

arXiv2025

Matrix-Free Two-to-Infinity and One-to-Two Norms Estimation

In this paper, we propose new randomized algorithms for estimating the two-to-infinity and one-to-two norms in a matrix-free setting, using only matrix-vector multiplications. Our methods are based on appropriate modifications of Hutchinson's diagonal estimator and its Hutch++ version. We provide oracle complexity bounds for both modifications. We further illustrate the practical utility of our algorithms for Jacobian-based regularization in deep neural network training on image classification tasks. We also demonstrate that our methodology can be applied to mitigate the effect of adversarial attacks in the domain of recommender systems.

cs.LG↗

arXiv2025

LoRA meets Riemannion: Muon Optimizer for Parametrization-independent Low-Rank Adapters

This work presents a novel, fully Riemannian framework for Low-Rank Adaptation (LoRA) that geometrically treats low-rank adapters by optimizing them directly on the fixed-rank manifold. This formulation eliminates the parametrization ambiguity present in standard Euclidean optimizers. Our framework integrates three key components to achieve this: (1) we derive Riemannion, a new Riemannian optimizer on the fixed-rank matrix manifold that generalizes the recently proposed Muon optimizer; (2) we develop a Riemannian gradient-informed LoRA initialization, and (3) we provide an efficient implementation without prominent overhead that uses automatic differentiation to compute arising geometric operations while adhering to best practices in numerical linear algebra. Comprehensive experimental results on both LLM and diffusion model architectures demonstrate that our approach yields consistent and noticeable improvements in convergence speed and final task performance over both standard LoRA and its state-of-the-art modifications.

cs.LG↗

arXiv2025

COALA: Numerically Stable and Efficient Framework for Context-Aware Low-Rank Approximation

Recent studies suggest that context-aware low-rank approximation is a useful tool for compression and fine-tuning of modern large-scale neural networks. In this type of approximation, a norm is weighted by a matrix of input activations, significantly improving metrics over the unweighted case. Nevertheless, existing methods for neural networks suffer from numerical instabilities due to their reliance on classical formulas involving explicit Gram matrix computation and their subsequent inversion. We demonstrate that this can degrade the approximation quality or cause numerically singular matrices. To address these limitations, we propose a novel inversion-free regularized framework that is based entirely on stable decompositions and overcomes the numerical pitfalls of prior art. Our method can handle possible challenging scenarios: (1) when calibration matrices exceed GPU memory capacity, (2) when input activation matrices are nearly singular, and even (3) when insufficient data prevents unique approximation. For the latter, we prove that our solution converges to a desired approximation and derive explicit error bounds.

cs.LG↗

arXiv2025

Pay Attention to Attention Distribution: A New Local Lipschitz Bound for Transformers

We introduce a novel upper bound on the local Lipschitz constant of the dot-product self-attention block showing its dependence on the attention map distributions. The proposed bound is not only tighter than the prior art, but for the first time, reveals how the distribution of attention probabilities shapes the local Lipschitz constant of the self-attention block. The theoretical basis of the proposed upper bound lies in the refined closed-form upper bounds on singular values of the Jacobian of softmax function. Leveraging these theoretical insights, we introduce JaSMin (Jacobian Softmax norm Minimization), a lightweight regularizer that directly controls the local Lipschitz constant of each block and, consequently, the entire model. Additionally, we discuss how the nature of the attention map distribution contributes to the gradient dynamics and, consequently, transformer training stability.

cs.LG↗

arXiv2025

On the Upper Bounds for the Matrix Spectral Norm

We consider the problem of estimating the spectral norm of a matrix using only matrix-vector products. We propose a new Counterbalance estimator that provides upper bounds on the norm and derive probabilistic guarantees on its underestimation. Compared to standard approaches such as the power method, the proposed estimator produces significantly tighter upper bounds in both synthetic and real-world settings. Our method is especially effective for matrices with fast-decaying spectra, such as those arising in deep learning and inverse problems.

math.NA↗

arXiv2025

Accelerating Newton-Schulz Iteration for Orthogonalization via Chebyshev-type Polynomials

The problem of computing optimal orthogonal approximation to a given matrix has attracted growing interest in machine learning. Notable applications include the recent Muon optimizer or Riemannian optimization on the Stiefel manifold. Among existing approaches, the Newton-Schulz iteration has emerged as a particularly effective solution, as it relies solely on matrix multiplications and thus achieves high computational efficiency on GPU hardware. Despite its efficiency, the method has inherent limitations - its coefficients are fixed and thus not optimized for a given matrix. In this paper we address this issue by proposing a Chebyshev-optimized version of Newton-Schulz (CANS). Based on the Chebyshev's alternance theorem, we theoretically derive optimal coefficients for the 3-rd order Newton-Schulz iteration and apply a Remez algorithm to compute optimal higher-degree polynomials. We leverage these polynomials to construct controlled approximate orthogonalization schemes, which is of interest in deep learning applications. Practically, we demonstrate the method's effectiveness in two key applications: orthogonalization in the Muon optimizer, and providing an efficient retraction alternative for Riemannian optimization on the Stiefel manifold.

math.NA↗

arXiv2025

ProcrustesGPT: Compressing LLMs with Structured Matrices and Orthogonal Transformations

Large language models (LLMs) demonstrate impressive results in natural language processing tasks but require a significant amount of computational and memory resources. Structured matrix representations are a promising way for reducing the number of parameters of these models. However, it seems unrealistic to expect that weight matrices of pretrained models can be accurately represented by structured matrices without any fine-tuning. To overcome this issue, we utilize the fact that LLM output is invariant under certain orthogonal transformations of weight matrices. This insight can be leveraged to identify transformations that significantly improve the compressibility of weights within structured classes. The proposed approach is applicable to various types of structured matrices that support efficient projection operations. Code is available at https://github.com/GrishKate/ProcrustesGPT

cs.CL↗

arXiv2025

Knowledge Graph Completion with Mixed Geometry Tensor Factorization

In this paper, we propose a new geometric approach for knowledge graph completion via low rank tensor approximation. We augment a pretrained and well-established Euclidean model based on a Tucker tensor decomposition with a novel hyperbolic interaction term. This correction enables more nuanced capturing of distributional properties in data better aligned with real-world knowledge graphs. By combining two geometries together, our approach improves expressivity of the resulting model achieving new state-of-the-art link prediction accuracy with a significantly lower number of parameters compared to the previous Euclidean and hyperbolic models.

cs.LG↗