Research output: Chapter in Book/Report/Conference proceeding › Conference contribution › Research › peer-review
Regularization of Machine Learning and Linear Algebra. / Liu, Shuang; Kabanikhin, Sergey Igorevich; Strijhak, Sergei Vladimirovich.
CSAI 2024 - Proceedings of 2024 8th International Conference on Computer Science and Artificial Intelligence. ed. / Chen Xiangqun; Fang Juan. Association for Computing Machinery, 2025. p. 327-332.Research output: Chapter in Book/Report/Conference proceeding › Conference contribution › Research › peer-review
}
TY - GEN
T1 - Regularization of Machine Learning and Linear Algebra
AU - Liu, Shuang
AU - Kabanikhin, Sergey Igorevich
AU - Strijhak, Sergei Vladimirovich
N1 - Conference code: 8
PY - 2025/2/15
Y1 - 2025/2/15
N2 - This paper explores the connection between systems of linear algebraic equations (SLAE) and machine learning methods, including regularization techniques, to establish a more novel neural network model based on linear neural networks. The goal is to construct a weight matrix for the neural network, which, by simulating the process of finding pseudo-solutions to SLAE, can generate the optimal answer for any input data. In this new neural network model, linear operations are performed first, followed by nonlinear operations, ultimately yielding an optimized weight matrix that serves as the pseudo-solution to the SLAE. The paper demonstrates how linear neural networks can be simplified to SLAE, how adding nonlinear layers to the linear neural network model can improve accuracy, and how machine learning methods can be used to find pseudo-solutions to SLAE.
AB - This paper explores the connection between systems of linear algebraic equations (SLAE) and machine learning methods, including regularization techniques, to establish a more novel neural network model based on linear neural networks. The goal is to construct a weight matrix for the neural network, which, by simulating the process of finding pseudo-solutions to SLAE, can generate the optimal answer for any input data. In this new neural network model, linear operations are performed first, followed by nonlinear operations, ultimately yielding an optimized weight matrix that serves as the pseudo-solution to the SLAE. The paper demonstrates how linear neural networks can be simplified to SLAE, how adding nonlinear layers to the linear neural network model can improve accuracy, and how machine learning methods can be used to find pseudo-solutions to SLAE.
KW - linear neural network
KW - machine learning
KW - system of linear algebraic equations
UR - https://www.scopus.com/record/display.uri?eid=2-s2.0-85219594269&origin=inward&txGid=4fbd80970b131287e75d4dfd57829bfa
UR - https://www.mendeley.com/catalogue/bef1d79a-e7c9-3b7e-a84d-a666173c46a0/
U2 - 10.1145/3709026.3709071
DO - 10.1145/3709026.3709071
M3 - Conference contribution
SN - 9798400718182
SP - 327
EP - 332
BT - CSAI 2024 - Proceedings of 2024 8th International Conference on Computer Science and Artificial Intelligence
A2 - Xiangqun, Chen
A2 - Juan, Fang
PB - Association for Computing Machinery
T2 - 8th International Conference on Computer Science and Artificial Intelligence
Y2 - 6 December 2024 through 8 December 2024
ER -
ID: 64991087