Understanding Scholarly Neural Network System Diagrams Through Application of VisDNA

We utilise VisDNA as a tool for understanding neural network system architecture diagrams. Through examples from scholarly proceedings, we find that the application of the framework to this ecological and complex domain is effective for reflecting on these diagrams. We argue for additional vocabulary to describe semiotic variability and internal inconsistency or misuse of visual encoding principles in diagrams. Further, for application to system diagrams, we propose the addition of “Grouping by Object” as a new visual encoding principle, and “Emphasising” as a new visual encoding type.