VIS4AI drives the development of ML models through three main tasks [44, 154, 270], model understanding, model diagnosis, and model steering. Model understanding aims to explain the rationale behind ML model predictions and their inner workings, and make these complex models at least partially understandable. Model diagnosis is the process of identifying and addressing defects or issues within ML models that do not converge or do not achieve acceptable performance. Model steering is a method of interactively incorporating expert knowledge and expertise into the improvement and refinement process of an ML model, which is achieved by combining a set of rich user interactions with techniques such as semi-supervised learning or active learning. Various visualization techniques, such as node-link diagrams, scatterplots, and parallel coordinate plots, have been adopted to support these three tasks. Figure 4.1 shows the connections between ML models, the three VIS4AI tasks, and different visualization techniques.