Intentional Ambiguity: Parts-of-a-Whole

Among the questions we ask, is how much information is needed to identify a part or reconstruct a scene. The related question is how we can maximize our understanding of the parts to generalize knowledge? This can be likened to how children learn, to how they are able to take diverse representations of birds during early childhood and be able to see birds in all their inter and intra-variability in color, shape, size and presentation format from cartoon characters to an eagle soaring in the sky - each in their complexity are simply BOORDS.

In our research, we seek to understand how spatially separated image parts in content-rich, stable and multi-stable image sets may be able to provide a unique opportunity to develop "logic leaping" training sets which can holistically analyze images while also considering its parts. This is similar to distractors and attractors we humans must contend with when analyzing scenes. Image sets can contain flanking content which can be be as adversarial, or utilize spatially separated image sections allowing deep learning networks to resolve the ambiguities, reconstruct the image scene and essentially solve what amounts to an image puzzle.