Results 291 to 300 of about 12,418,444 (305)

LoRA-Pro: Are Low-Rank Adapters Properly Optimized?

International Conference on Learning Representations
Low-rank adaptation, also known as LoRA, has emerged as a prominent method for parameter-efficient fine-tuning of foundation models. Despite its computational efficiency, LoRA still yields inferior performance compared to full fine-tuning. In this paper,
Zhengbo Wang, Jian Liang
semanticscholar   +1 more source

Mixture-of-Subspaces in Low-Rank Adaptation

Conference on Empirical Methods in Natural Language Processing
In this paper, we introduce a subspace-inspired Low-Rank Adaptation (LoRA) method, which is computationally efficient, easy to implement, and readily applicable to large language, multimodal, and diffusion models. Initially, we equivalently decompose the
Taiqiang Wu, Jiahao Wang, Zhe Zhao, Ngai Wong   +3 more
semanticscholar   +1 more source

Low‐rank revealing QR factorizations

Numerical Linear Algebra with Applications, 1994
AbstractRank revealing factorizations are used extensively in signal processing in connection with, for example, linear prediction and signal subspace algorithms. We present an algorithm for computing rank revealing QR factorizations of low‐rank matrices.
Chan, Tony F., Hansen, Per Christian
openaire   +1 more source

Compressing Large Language Models using Low Rank and Low Precision Decomposition

Neural Information Processing Systems
The prohibitive sizes of Large Language Models (LLMs) today make it difficult to deploy them on memory-constrained edge devices. This work introduces $\rm CALDERA$ -- a new post-training LLM compression algorithm that harnesses the inherent low-rank ...
R. Saha, Naomi Sagan, Varun Srivastava, Andrea J. Goldsmith, Mert Pilanci   +4 more
semanticscholar   +1 more source

Low-rank physical model recovery from low-rank signal approximation

2017 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), 2017
This work presents a mathematical approach for recovering a physical model from a low-rank approximation of measured data obtained via the singular value decomposition (SVD). The general form of a low-rank physical model of the data is often known, so the presented approach learns the proper rotation and scaling matrices from the singular vectors and ...
Charles Ethan Hayes, James H. McClellan, Waymond R. Scott   +2 more
openaire   +1 more source

Robust Kernel Low-Rank Representation

IEEE Transactions on Neural Networks and Learning Systems, 2016
Recently, low-rank representation (LRR) has shown promising performance in many real-world applications such as face clustering. However, LRR may not achieve satisfactory results when dealing with the data from nonlinear subspaces, since it is originally designed to handle the data from linear subspaces in the input space.
Shijie, Xiao, Mingkui, Tan, Dong, Xu, Zhao Yang, Dong   +3 more
openaire   +2 more sources

Low-rank revealing UTV decompositions

Numerical Algorithms, 1997
Much attention has been paid to the \(UTV\) decompositions of a high-rank matrix, however, only a little to the low-rank case. Low-rank matrices arise when a small number of parameters suffices to describe a system. The high-rank revealing algorithms are not suited for such problems, and hence there is a need for algorithms which handle the low-rank ...
Fierro, Ricardo D., Hansen, Per Christian   +1 more
openaire   +2 more sources

Low Rank Solution of Lyapunov Equations

SIAM Journal on Matrix Analysis and Applications, 2002
The Cholesky factor-alternating direction implicit algorithm is presented to compute a low rank approximation to the solution \(X\) of the Lyapunov equation \(AX+XA^T=-BB^T\) with large matrix \(A\) and right hand side of low rank. The algorithm requires only matrix-vector products and linear solvers.
Li, Jing-Rebecca, White, Jacob
openaire   +2 more sources

Low-Rank Dynamics

2014
This note reviews differential equations on manifolds of matrices or tensors of low rank. They serve to approximate, in a low-rank format, large time-dependent matrices and tensors that are either given explicitly via their increments or are unknown solutions of differential equations.
openaire   +1 more source

Low Rank Approximations

2019
A principal components analysis models high dimensional data points with an accurate, low dimensional, model. Now form a data matrix from the approximate points. This data matrix must have low rank (because the model is low dimensional) and it must be close to the original data matrix (because the model is accurate). This suggests modelling data with a
openaire   +1 more source