A Comprehensive Survey on Data Augmentation

2024-05-15

arXiv.org

AI Summary

Existing surveys are limited to unimodal or operation-centric taxonomies, lacking a unified conceptualization of cross-modal data augmentation. To address this, we propose the first modality-agnostic classification framework centered on *intrinsic data relationships*, systematically organizing augmentation techniques across five modalities—images, text, speech, time series, and graph-structured data—according to sample granularity (single-sample, sample-pair, and population-level). Moving beyond conventional modality- or transformation-based categorizations, our framework adopts a *data-centric perspective* to uncover the fundamental principle of *relationship-driven augmentation*. Through inductive analysis, cross-modal comparison, and empirical validation, we construct a comprehensive, three-tiered taxonomy grounded in relational dimensions—semantic, structural, and distributional—which spans all modalities. This taxonomy significantly enhances method comparability, cross-modal transferability, and theoretical coherence in data augmentation research.

Technology Category

• Machine Learning: Multimodal Learning
• Computer Vision: Multi-modal Vision
• Data Mining & Knowledge Management: Mining of Visual, Multimedia & Multimodal Data

Application Category

• Web Mining and Content Analysis: Mining multimedia, multimodal, multilingual, cross-lingual Web data
• Search and Retrieval-Augmented AI: Retrieval-Augmented Generation (RAG) and multi-modal RAG
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs

Abstract

Data augmentation is a series of techniques that generate high-quality artificial data by manipulating existing data samples. By leveraging data augmentation techniques, AI models can achieve significantly improved applicability in tasks involving scarce or imbalanced datasets, thereby substantially enhancing AI models' generalization capabilities. Existing literature surveys only focus on a certain type of specific modality data, and categorize these methods from modality-specific and operation-centric perspectives, which lacks a consistent summary of data augmentation methods across multiple modalities and limits the comprehension of how existing data samples serve the data augmentation process. To bridge this gap, we propose a more enlightening taxonomy that encompasses data augmentation techniques for different common data modalities. Specifically, from a data-centric perspective, this survey proposes a modality-independent taxonomy by investigating how to take advantage of the intrinsic relationship between data samples, including single-wise, pair-wise, and population-wise sample data augmentation methods. Additionally, we categorize data augmentation methods across five data modalities through a unified inductive approach.

Problem

• Survey lacks consistent multi-modal data augmentation summary.
• Existing taxonomies are modality-specific, not data-centric.
• Proposes unified taxonomy for five data modalities.

Innovation

• Proposes modality-independent taxonomy for data augmentation
• Leverages intrinsic relationships between data samples
• Categorizes methods across five data modalities