Learn2Mix: Training Neural Networks Using Adaptive Data Integration

Shyam Venkatasubramanian (Duke University) · Vahid Tarokh (Duke University)

adaptive adjustment benchmark datasetsclass proportionsclassification tasks empirical evaluations error ratesimbalanced classeslearn2mixlimited training resourcesmodel convergenceneural network trainingreconstruction tasksregression tasks resource-constrained environments theoretical analysis training strategy

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.