Fig. 1: Model-centric AI vs. data-centric AI.
From: Rethinking deep learning in bioimaging through a data centric lens
Suppose we roughly divide the developmental process of a project into three phases: preparation, development, and deployment (marked by three different colors). a, b illustrate a model-centric approach and a data-centric approach, respectively. The relative sizes of the Model block and the Data block indicate the amount of effort invested in the corresponding parts. In the end, the model-centric approach delivers a powerful model with an excellent evaluation score on seen data. For the data-centric approach, the deliverables include a high-quality dataset (re-usable for future models) with a suitable model that can produce reliable analysis in practice, even on new data. A prototypical BioData-Centric framework for AI-based bioimage analysis is outlined in (c).
