Influence-guided Data Augmentation for Neural Tensor Completion

TL;DR

This paper introduces DAIN, a data augmentation framework that improves neural tensor completion accuracy by leveraging influence functions to identify and sample important tensor cells, outperforming existing methods.

Contribution

The paper presents a novel data augmentation method for tensors that enhances neural tensor completion, addressing overfitting and sparsity issues in real-world data.

Findings

DAIN outperforms all baseline augmentation methods in accuracy.

DAIN scales near linearly to large datasets.

Ablation studies confirm the effectiveness of each component.

Abstract

How can we predict missing values in multi-dimensional data (or tensors) more accurately? The task of tensor completion is crucial in many applications such as personalized recommendation, image and video restoration, and link prediction in social networks. Many tensor factorization and neural network-based tensor completion algorithms have been developed to predict missing entries in partially observed tensors. However, they can produce inaccurate estimations as real-world tensors are very sparse, and these methods tend to overfit on the small amount of data. Here, we overcome these shortcomings by presenting a data augmentation technique for tensors. In this paper, we propose DAIN, a general data augmentation framework that enhances the prediction accuracy of neural tensor completion methods. Specifically, DAIN first trains a neural model and finds tensor cell importances with influence functions. After that, DAIN aggregates the cell importance to calculate the importance of each entity (i.e., an index of a dimension). Finally, DAIN augments the tensor by weighted sampling of entity importances and a value predictor. Extensive experimental results show that DAIN outperforms all data augmentation baselines in terms of enhancing imputation accuracy of neural tensor completion on four diverse real-world tensors. Ablation studies of DAIN substantiate the effectiveness of each component of DAIN. Furthermore, we show that DAIN scales near linearly to large datasets.