# Data-centric Artificial Intelligence: A Survey

**Authors:** Daochen Zha, Zaid Pervaiz Bhat, Kwei-Herng Lai, Fan Yang, Zhimeng Jiang, Shaochen Zhong, Xia Hu  
**arXiv:** [2303.10158](https://arxiv.org/abs/2303.10158)  
**Date:** 2023-03-17  
**Summary page:** https://www.summarizepaper.com/en/paper/2303.10158v1/

## Key points

- Artificial Intelligence (AI) has revolutionized various domains
- Availability of abundant and high-quality data is crucial for training machine learning models
- Data-centric AI has gained prominence in recent years
- Three key goals of data-centric AI: training data development, inference data development, and data maintenance
- Various methods are employed to enhance the quality and quantity of data in each stage of the data lifecycle
- Challenges faced in achieving effective data-centric AI from automation and collaboration perspectives
- Benchmarks for different tasks are tabulated, providing a resource for evaluating performance
- The survey offers a global view across multiple stages of the data lifecycle in a spectrum of tasks related to AI
- It aims to help readers gain a comprehensive understanding of this field and provide techniques and research ideas for engineering robust AI systems
- An accompanying list of resources on data-centric AI is regularly updated on https://github.com/daochenzha/data-centric-AI.
- The survey serves as an authoritative reference for researchers and practitioners seeking insights into leveraging high-quality data for advancing AI capabilities.

## Summary

Artificial Intelligence (AI) has revolutionized various domains, and its success can be attributed to the availability of abundant and high-quality data for training machine learning models. In recent years, the concept of data-centric AI has gained prominence as researchers and practitioners recognize the crucial role of data in AI systems. This comprehensive survey by Daochen Zha et al. explores the necessity of data-centric AI and provides a holistic view of three key goals: training data development, inference data development, and data maintenance. The survey delves into various methods employed to enhance the quality and quantity of data in each stage of the data lifecycle. It also discusses existing literature from automation and collaboration perspectives, highlighting the challenges faced in achieving effective data-centric AI. Additionally, benchmarks for different tasks are tabulated, providing a valuable resource for evaluating performance. This survey stands out as one of the first to offer a global view across multiple stages of the data lifecycle in a spectrum of tasks related to AI. Its aim is to help readers gain a comprehensive understanding of this field while equipping them with techniques and research ideas for systematically engineering data to build robust AI systems. To further support readers' exploration, an accompanying list of resources on data-centric AI is regularly updated on https://github.com/daochenzha/data-centric-AI. With 38 pages, 6 figures, and 5 tables, this survey serves as an authoritative reference for researchers and practitioners seeking insights into leveraging high-quality data for advancing AI capabilities.