Learn2Mix: Training Neural Networks Using Adaptive Data Integration
This addresses the need for efficient training in environments with limited resources and imbalanced data, though it appears incremental as an adaptive extension of classical training methods.
The paper tackles the problem of accelerating neural network convergence in resource-constrained settings by introducing learn2mix, a training strategy that adaptively adjusts class proportions in batches based on error rates, leading to faster convergence than existing methods across classification, regression, and reconstruction tasks.
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.