Saved in:
Bibliographic Details
Main Author: Wang, Xi
Format: Preprint
Published: 2025
Subjects:
Online Access:https://arxiv.org/abs/2503.11965
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866908272875798528
author Wang, Xi
author_facet Wang, Xi
contents We introduce a novel framework for learning in neural networks by decomposing each neuron's weight vector into two distinct parts, $W_1$ and $W_2$, thereby modeling contrastive information directly at the neuron level. Traditional gradient descent stores both positive (target) and negative (non-target) feature information in a single weight vector, often obscuring fine-grained distinctions. Our approach, by contrast, maintains separate updates for target and non-target features, ultimately forming a single effective weight $W = W_1 - W_2$ that is more robust to noise and class imbalance. Experimental results on both regression (California Housing, Wine Quality) and classification (MNIST, Fashion-MNIST, CIFAR-10) tasks suggest that this decomposition enhances generalization and resists overfitting, especially when training data are sparse or noisy. Crucially, the inference complexity remains the same as in the standard $WX + \text{bias}$ setup, offering a practical solution for improved learning without additional inference-time overhead.
format Preprint
id arxiv_https___arxiv_org_abs_2503_11965
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Revisiting Gradient Descent: A Dual-Weight Method for Improved Learning
Wang, Xi
Machine Learning
We introduce a novel framework for learning in neural networks by decomposing each neuron's weight vector into two distinct parts, $W_1$ and $W_2$, thereby modeling contrastive information directly at the neuron level. Traditional gradient descent stores both positive (target) and negative (non-target) feature information in a single weight vector, often obscuring fine-grained distinctions. Our approach, by contrast, maintains separate updates for target and non-target features, ultimately forming a single effective weight $W = W_1 - W_2$ that is more robust to noise and class imbalance. Experimental results on both regression (California Housing, Wine Quality) and classification (MNIST, Fashion-MNIST, CIFAR-10) tasks suggest that this decomposition enhances generalization and resists overfitting, especially when training data are sparse or noisy. Crucially, the inference complexity remains the same as in the standard $WX + \text{bias}$ setup, offering a practical solution for improved learning without additional inference-time overhead.
title Revisiting Gradient Descent: A Dual-Weight Method for Improved Learning
topic Machine Learning
url https://arxiv.org/abs/2503.11965