?
Training a Tucker Model With Shared Factors: a Riemannian Optimization Approach
Ch. 238. P. 3304–3312.
Language:
English
Vladimir Bogachev, Aletov V., Alexander Molozhavenko et al., , in: The Fourteenth International Conference on Learning Representations (ICLR 2026).: ICLR, 2026. Ch. 20503 P. 1–26.
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 ...
Added: April 29, 2026
Alexander Molozhavenko, Rakhuba M., Computational and Applied Mathematics 2026 Vol. 45 No. 6 Article 221
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 ...
Added: December 22, 2025
Novikov A., Rakhuba M., Oseledets I., SIAM Journal of Scientific Computing 2022 Vol. 44 No. 2 P. A843–A869
In scientific computing and machine learning applications, matrices and more general multidimensional arrays (tensors) can often be approximated with the help of low-rank decompositions. Since matrices and tensors of fixed rank form smooth Riemannian manifolds, one of the popular tools for finding low-rank approximations is to use Riemannian optimization. Nevertheless, efficient implementation of Riemannian gradients ...
Added: October 31, 2021
IEEE, 2020.
Dimensionality reduction problem is stated as finding a mapping f:X ∈ R m → Z ∈ R n , where ≪ m while preserving some relevant properties of the data. We formulate topology-preserving dimensionality reduction as finding the optimal orthogonal projection to the lower-dimensional subspace which minimizes discrepancy between persistent diagrams of the original data and the projection. This ...
Added: October 14, 2021
Rakhuba M., Oseledets I., SIAM Journal of Scientific Computing 2018 Vol. 40 No. 2 P. A1149–A1170
In this work we generalize the Jacobi--Davidson method to the case when the eigenvector can be reshaped into a low-rank matrix. In this setting the proposed method inherits the advantages of the original Jacobi--Davidson method, has lower complexity, and requires less storage. We also introduce a low-rank version of the Rayleigh quotient iteration which naturally ...
Added: October 20, 2020
Rakhuba Maxim, Novikov A., Oseledets I., Journal of Computational Physics 2019 Vol. 396 P. 718–737
Such problems as computation of spectra of spin chains and vibrational spectra of molecules can be written as high-dimensional eigenvalue problems, i.e., when the eigenvector can be naturally represented as a multidimensional tensor. Tensor methods have proven to be an efficient tool for the approximation of solutions of high-dimensional eigenvalue problems, however, their performance deteriorates quickly ...
Added: October 19, 2020
Novikov A., Trofimov M., Oseledets I., / Series stat :: arxiv :: Cornell University "stat :: arxiv :: Cornell University". 2017.
Modeling interactions between features improves the performance of machine learning solutions in many domains (e.g. recommender systems or sentiment analysis). In this paper, we introduce Exponential Machines (ExM), a predictor that models all interactions of every order. The key idea is to represent an exponentially large tensor of parameters in a factorized format called Tensor ...
Added: September 19, 2016