Kandinsky Patterns

Kandinsky Figures and Patterns are mathematically describable, simple self-contained hence controllable test data sets for the development, validation training of explainability in artificial intelligence. Whilst have these computationally manageable properties, they at same time easily distinguishable from human observers. Consequently, controlled patterns can be described by both humans computers. We define a Pattern as set Figures, where each figure an "infallible authority" defines that belongs to Pattern. With this principle we build automatic interpretability context learning. In paper describe basic idea some underlying principles provide Github repository invite international machine learning research community challenge experiment with our expand thus make progress field explainable AI contribute upcoming causability.