Saved in:
Bibliographic Details
Main Authors: Chen, Qian, Yang, Linxin, Wang, Akang, Luo, Xiaodong, Zhang, Yin
Format: Preprint
Published: 2025
Subjects:
Online Access:https://arxiv.org/abs/2510.03276
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866909823581290496
author Chen, Qian
Yang, Linxin
Wang, Akang
Luo, Xiaodong
Zhang, Yin
author_facet Chen, Qian
Yang, Linxin
Wang, Akang
Luo, Xiaodong
Zhang, Yin
contents The combination of linear transformations and non-linear activation functions forms the foundation of most modern deep neural networks, enabling them to approximate highly complex functions. This paper explores the introduction of quadratic transformations to further increase nonlinearity in neural networks, with the aim of enhancing the performance of existing architectures. To reduce parameter complexity and computational complexity, we propose a lightweight quadratic enhancer that uses low-rankness, weight sharing, and sparsification techniques. For a fixed architecture, the proposed approach introduces quadratic interactions between features at every layer, while only adding negligible amounts of additional model parameters and forward computations. We conduct a set of proof-of-concept experiments for the proposed method across three tasks: image classification, text classification, and fine-tuning large-language models. In all tasks, the proposed approach demonstrates clear and substantial performance gains.
format Preprint
id arxiv_https___arxiv_org_abs_2510_03276
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle QuadEnhancer: Leveraging Quadratic Transformations to Enhance Deep Neural Networks
Chen, Qian
Yang, Linxin
Wang, Akang
Luo, Xiaodong
Zhang, Yin
Machine Learning
Artificial Intelligence
The combination of linear transformations and non-linear activation functions forms the foundation of most modern deep neural networks, enabling them to approximate highly complex functions. This paper explores the introduction of quadratic transformations to further increase nonlinearity in neural networks, with the aim of enhancing the performance of existing architectures. To reduce parameter complexity and computational complexity, we propose a lightweight quadratic enhancer that uses low-rankness, weight sharing, and sparsification techniques. For a fixed architecture, the proposed approach introduces quadratic interactions between features at every layer, while only adding negligible amounts of additional model parameters and forward computations. We conduct a set of proof-of-concept experiments for the proposed method across three tasks: image classification, text classification, and fine-tuning large-language models. In all tasks, the proposed approach demonstrates clear and substantial performance gains.
title QuadEnhancer: Leveraging Quadratic Transformations to Enhance Deep Neural Networks
topic Machine Learning
Artificial Intelligence
url https://arxiv.org/abs/2510.03276