This site aims to provide resources to support the creation of good diagrams for AI Systems.
Some diagrams suck.
In currently unpublished work, it has been shown that there is a problem with the many different types of diagram being used to represent AI systems.
Inspired by this jungle of diagrams, we are conducting experiments to discover which specific attributes of AI diagrams are helpful to researchers. We hope that by providing a framework to construct diagrams in a more empirically-grounded way will help ensure that, as a mapping between spaces, they are complete, meaningful, unambiguous and correct.
What is DIAL?
We started out as a standardisation effort (contact us to get involved!) and quickly found there was a lack of evidence around representational choices made in diagrams in general, let alone with the more complex requirements of AI systems (like changing data representations, layers, etc.). Now, we’re focusing more on experimental evidence than on a standardised language, and hope this will lay the groundwork for a future DIAL v1.0, which we’d love you to be a part of.