Learn2Mix: Training Neural Networks Using Adaptive Data Integration

Shyam Venkatasubramanian, Vahid Tarokh
12/21/2024
cs.LGstat.ML

No official code found. Best implementation: shyamven/Learn2Mix (0 stars, last commit 2025-10-23).

Abstract

Accelerating model convergence in resource-constrained environments is essential for fast and efficient neural network training. This work presents learn2mix, a new training strategy that adaptively adjusts class proportions within batches, focusing on classes with higher error rates. Unlike classical training methods that use static class proportions, learn2mix continually adapts class proportions during training, leading to faster convergence. Empirical evaluations on benchmark datasets show that neural networks trained with learn2mix converge faster than those trained with existing approaches, achieving improved results for classification, regression, and reconstruction tasks under limited training resources and with imbalanced classes. Our empirical findings are supported by theoretical analysis.

View on arXivView PDF

Code Implementations(1)

shyamven/Learn2MixCommunity100%

Learn2Mix: Training Neural Networks for Adaptive Data Integration (NeurIPS '25)

00Python, Jupyter NotebookDec 21, 20241 years ago

Similar Papers

Adaptive Optimization Algorithms for Machine Learning

81% match2023

Adaptive Class Emergence Training: Enhancing Neural Network Stability and Generalization through Progressive Target Evolution

80% match2024

DIVEBATCH: Accelerating Model Training Through Gradient-Diversity Aware Batch Size Adaptation

80% match2025

Dynamic Batch Adaptation

79% match2022

Accelerated Training via Incrementally Growing Neural Networks using Variance Transfer and Learning Rate Adaptation

79% match2023

Data Uniformity Improves Training Efficiency and More, with a Convergence Framework Beyond the NTK Regime

79% match1 repo62025

Cite this paper

@article{venkatasubramanian2024learnmix,
  title  = {Learn2Mix: Training Neural Networks Using Adaptive Data Integration},
  author = {Shyam Venkatasubramanian and Vahid Tarokh},
  year   = {2024},
  eprint = {2412.16482},
  archivePrefix = {arXiv},
  url    = {http://arxiv.org/abs/2412.16482v3},
  journal = {NEURIPS 2024 2024}
}

Add a badge to your README

Shows whether official code is linked for this paper.

Code of Paper badge
[![Code of Paper](https://codeofpaper.com/badge/2412.16482.svg)](https://codeofpaper.com/paper/2412.16482)

Discussion