.png){kind=icon}
返回
A Simple Alternating Framework for ReLU-Based Nonlinear Tensor Decomposition
.png){kind=icon}
PRE
AI
DOI：10.1109/LSP.2025.3649594.png)
.png){kind=icon}
原文求助.png){kind=icon}
分享.png){kind=icon}
收藏
摘要
En 中文
We study a nonlinear low–rank tensor decomposition model in which the observed nonnegative tensor $\mathcal {M}$ is generated by rectifying a latent low–rank tensor $\mathcal {X}$ element-wise via the ReLU function, i. e., $\mathcal {M}\approx \max (0,\mathcal {X})$. This paradigm is motivated by data that are sparse and nonnegative in the ambientspace but admit a much more compact representation before rectification. To solve the resulting constrained nonconvex model, we introduce a simple alternating minimization framework, which is denoted by NTD-tSVD. We also briefly analyze the computational complexity and convergence of the proposed algorithm. Extensive experiments on synthetic and real datasets demonstrate consistent improvements over direct linear low–rank approximations, yielding a markedly lower reconstruction discrepancy in the rectified domain, especially when the observations are highly sparse.
Keyword:
Alternating minimization
low-rank
nonlinear tensor decomposition
期刊
I
IF:
3.9
论文数:
833
被引数:
0
机构
引用论文
An Accelerated Alternating Partial Bregman Algorithm for ReLU-based Matrix Decomposition一种加速交替部分Bregman算法用于基于ReLU的矩阵分解
PREAI
分享收藏CP decomposition for tensors via alternating least squares with QR decomposition张量CP分解的交替最小二乘法与QR分解结合
OAAI
OAAI分享收藏
Faster nonconvex low-rank matrix learning for image low-level and high-level vision: A unified framework.png)
INFORMATION FUSION
INFORMATION FUSIONIF15.5
PREAI
分享收藏
查看更多内容