There has been a discrepancy between academic research and industrial applications. Academic research weighs more on developing new models, but industrial application weighs more on the data. Open data such as ImageNet, KITTI, and MNIST has been at the core of AI research in the last several decades. With the rise of open data, more researchers began to realize the importance of data in AI development. Industry expert Andrew Ng and many other developers are advocating for the transition from Model-centric AI to Data-centric AI development.


In this talk, we will discuss the rationale of Data-centric AI development from an academic perspective and explain some of the ways to improve data quality. We will also talk about some current pain points of open data and introduce Graviti Open Dataset --- our solution to these problems by showcasing a demo on its usage.