This project addresses the research problem of using machine learning (ML) for visualization (ML4Vis). We will develop visualization methods that employ a data-driven approach for feature specification and extraction in the context of multifield data. For this, we will especially use Neural Network architectures and formulate our visualization tasks as supervised learning problems. Our visualization presents identified characteristics of interest when applying trained models to the data, or uses the models for selecting and directly generating expressive visual representations for data analysis. This research directly tackles a fundamental challenge in visualization: meaningful data reduction and abstraction are required, yet solely relying on manual specification by the user is increasingly impractical considering the growing data size and complexity from simulations and experiments. The developed approaches will be applied together with SimTech collaborators to help them analyze, understand, and discover new characteristics in their data.