Leveraging Local Variation in Data: Sampling and Weighting Schemes for Supervised Deep Learning
This addresses the challenge of inefficient training data distribution for supervised learning, offering a scalable and cost-effective solution that enhances neural network accuracy in classification and regression across domains like image and text data.
The paper tackles the problem of improving neural network performance by focusing training on data regions where the target function is steep, proposing a method called Variance Based Samples Weighting (VBSW) that weights training points based on local label variance, which significantly boosts performance across various tasks and network architectures.
In the context of supervised learning of a function by a neural network, we claim and empirically verify that the neural network yields better results when the distribution of the data set focuses on regions where the function to learn is steep. We first traduce this assumption in a mathematically workable way using Taylor expansion and emphasize a new training distribution based on the derivatives of the function to learn. Then, theoretical derivations allow constructing a methodology that we call Variance Based Samples Weighting (VBSW). VBSW uses labels local variance to weight the training points. This methodology is general, scalable, cost-effective, and significantly increases the performances of a large class of neural networks for various classification and regression tasks on image, text, and multivariate data. We highlight its benefits with experiments involving neural networks from linear models to ResNet and Bert.