An Investigation into the Contribution of Individual Training Samples in Artificial Neural Network Learning with Mini-Batch Strategy.

Saved in:

Bibliographic Details
Title:	An Investigation into the Contribution of Individual Training Samples in Artificial Neural Network Learning with Mini-Batch Strategy.
Authors:	Hernadi, Julan¹ julan.hernadi@math.uad.ac.id
Source:	IAENG International Journal of Applied Mathematics. Mar2026, Vol. 56 Issue 3, p1185-1195. 11p.
Subjects:	Artificial neural networks, Mean square algorithms, Machine learning, Statistical sampling
Abstract:	The mini-batch strategy frequently represents a pragmatic compromise when training Artificial Neural Networks (ANN) based on several considerations, primarily its role in restraining the excessive volatility inherent in incremental learning and diminishing the substantial large data training for full-batch methods. In the context of mini-batch training, existing computer program toolboxes and libraries typically do not provide insights into the specific contribution of each individual sample to the resulting Mean Squared Error (MSE). Instead, achieving a small MSE is viewed as a collective success across the entire training dataset, and conversely, failing to minimize MSE is also perceived as a collective failure. In mini-batch strategies, all training data are typically uniformly involved in the learning process, grouped and processed within each epoch. However, there is no inherent requirement that the frequency of each individual training data point must be equal. This paper aims to investigate the ability of various training datasets to minimize MSE, with explicit focus on the contributions of individual training data points to accuracy. Concurrently, this research examined the correlation between the frequency of individual training data involvement and the resulting training accuracy. The mini-batch strategy algorithm proposed herein allows for varying frequencies of individual training data involvement across epochs. To support this investigation, a benchmark ANN architecture was established to assess training performance empirically. The findings suggest that while individual data points make varying contributions to accuracy, there is no apparent correlation between their frequency of use and their impact on accuracy. [ABSTRACT FROM AUTHOR]
	Copyright of IAENG International Journal of Applied Mathematics is the property of International Association of Engineers (IAENG) and its content may not be copied or emailed to multiple sites without the copyright holder's express written permission. Additionally, content may not be used with any artificial intelligence tools or machine learning technologies. However, users may print, download, or email articles for individual use. This abstract may be abridged. No warranty is given about the accuracy of the copy. Users should refer to the original published version of the material for the full abstract. (Copyright applies to all Abstracts.)
Database:	Engineering Source

Description
Abstract:	The mini-batch strategy frequently represents a pragmatic compromise when training Artificial Neural Networks (ANN) based on several considerations, primarily its role in restraining the excessive volatility inherent in incremental learning and diminishing the substantial large data training for full-batch methods. In the context of mini-batch training, existing computer program toolboxes and libraries typically do not provide insights into the specific contribution of each individual sample to the resulting Mean Squared Error (MSE). Instead, achieving a small MSE is viewed as a collective success across the entire training dataset, and conversely, failing to minimize MSE is also perceived as a collective failure. In mini-batch strategies, all training data are typically uniformly involved in the learning process, grouped and processed within each epoch. However, there is no inherent requirement that the frequency of each individual training data point must be equal. This paper aims to investigate the ability of various training datasets to minimize MSE, with explicit focus on the contributions of individual training data points to accuracy. Concurrently, this research examined the correlation between the frequency of individual training data involvement and the resulting training accuracy. The mini-batch strategy algorithm proposed herein allows for varying frequencies of individual training data involvement across epochs. To support this investigation, a benchmark ANN architecture was established to assess training performance empirically. The findings suggest that while individual data points make varying contributions to accuracy, there is no apparent correlation between their frequency of use and their impact on accuracy. [ABSTRACT FROM AUTHOR]
ISSN:	19929978