arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2412.16482cs.LGstat.ML
收藏

Learn2Mix:使用自适应数据整合训练神经网络

Learn2Mix: Training Neural Networks Using Adaptive Data Integration

Duke University(杜克大学)

Shyam Venkatasubramanian, Vahid Tarokh

AI总结:

本文提出 learn2mix 训练策略,通过自适应调整批次内类别比例并聚焦高错误率类别,在资源受限和类别不平衡条件下加速神经网络收敛,并在分类、回归和重建任务上取得更优结果。

摘要:

在资源受限的环境中加速模型收敛对于快速高效的神经网络训练至关重要。本文提出了一种新的训练策略 learn2mix,该策略在批次内自适应地调整类别比例,重点关注错误率较高的类别。与使用静态类别比例的传统训练方法不同,learn2mix 在训练过程中持续调整类别比例,从而实现更快的收敛。在基准数据集上的实证评估表明,使用 learn2mix 训练的神经网络比使用现有方法训练的神经网络收敛更快,在有限的训练资源和类别不平衡的情况下,在分类、回归和重建任务上均取得了改进的结果。我们的实证发现得到了理论分析的支持。

英文摘要:

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.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑