An Empirical Evaluation of Sketched SVD and its Application to Leverage Score Ordering
This work addresses the computational bottleneck of SVD in machine learning for practitioners dealing with large datasets, offering a practical, incremental improvement through empirical validation and application to data ordering.
The paper tackled the challenge of efficiently computing Singular Value Decomposition (SVD) on large-scale datasets by implementing and evaluating sketched SVD algorithms, showing faster performance and improved convergence in neural network training applications. It introduced Sketched Leverage Score Ordering, which automatically determines optimal data ordering without manual intervention, achieving better results across image classification, language sentiment analysis, and multi-modal tasks.
The power of randomized algorithms in numerical methods have led to fast solutions which use the Singular Value Decomposition (SVD) as a core routine. However, given the large data size of modern and the modest runtime of SVD, most practical algorithms would require some form of approximation, such as sketching, when running SVD. While these approximation methods satisfy many theoretical guarantees, we provide the first algorithmic implementations for sketch-and-solve SVD problems on real-world, large-scale datasets. We provide a comprehensive empirical evaluation of these algorithms and provide guidelines on how to ensure accurate deployment to real-world data. As an application of sketched SVD, we present Sketched Leverage Score Ordering, a technique for determining the ordering of data in the training of neural networks. Our technique is based on the distributed computation of leverage scores using random projections. These computed leverage scores provide a flexible and efficient method to determine the optimal ordering of training data without manual intervention or annotations. We present empirical results on an extensive set of experiments across image classification, language sentiment analysis, and multi-modal sentiment analysis. Our method is faster compared to standard randomized projection algorithms and shows improvements in convergence and results.