COPU: Recognizing Time Series' Heterogeneity In Stacked Neural Network
Keywords: Neural Network, Time Series, Regression, Nonlinear Modeling, Natural Gradient
TL;DR: Experimentally demonstrate the ineffectiveness of simply stacking neural networks for time series applications; and develop CEP, RR, and COPU as innovative solutions.
Abstract: Neural networks (NNs) have been widely studied in complex fields due to their remarkable capacity for nonlinear modeling.
However, in the realm of time series analysis, researches indicate that merely stacking NNs does not yield promising nonlinear modeling outputs and hinders model performance. Conventional NN architectures overemphasize homogeneous feature extraction, impeding the learning of diverse features and diminishing their nonlinear modeling capability. To address this gap, we propose the $\textbf{C}$ross-correlation Enhanced Approximated $\textbf{O}$rthogonal $\textbf{P}$rojection $\textbf{U}$nit (COPU) to quantify and augment the NN's nonlinear modeling capacity. COPU efficiently computes the local cross-correlation characteristics between features, amplifying heterogeneous components while compressing homogeneous ones. By reducing redundant information, COPU facilitates the learning of unique and independent features, thereby enhancing nonlinear modeling capability. Extensive experiments demonstrate that our method achieves superior performance across two real-world regression applications.
Primary Area: applications to computer vision, audio, language, and other modalities
Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.
Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide.
Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors’ identity.
No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.
Submission Number: 4144
Loading