1 Query images Result images Automatic generalization Dataset: wikiart images - https://www.wikiart.org/. No image pre-processing performed. No textual data used. Feature vectors generated automatically with deep learning. www.xyggy.com © 2019 

2 Walkthrough … www.xyggy.com © 2019 Query image is woman with long neck, small eyes and narrow face leaning to one side. Results (left to right) are in character with query image. Note: most art is unique and so identical similarity will be rare. 

3 www.xyggy.com © 2019 Add 2nd query image. Given one or more images, finds other relevant images in ranked order. Industry focus is on squeezing every pixel from image but metaphorically two eyes are better than one. 

4 www.xyggy.com © 2019 A modern art couple is 3rd query image. First result is duplicate of modern couple. The 2nd is in style of the long-necked woman. Remaining results are in style of modern couple. Portraits #3 and #8 are in character with first two query images. 

5 www.xyggy.com © 2019 The 4th query image is woman with long’ish neck and headscarf. Results include females with head-coverings, long necks and leaning heads. Result #8, in style of modern couple, has long neck and face. 

6 In a nutshell … www.xyggy.com © 2019 Given one or more items, Thingy finds other relevant items in ranked order 

7 What is a thing? www.xyggy.com © 2019 Items (“things”) can be images, audio, video, db records, documents, genes … in fact, all datatypes. Multi-modal? Yes, eventually. 

8 Dynamic AI www.xyggy.com © 2019 1. Automatic generalization – in realtime 2. Interactive – query with one or more items; add and remove items from query. Gives consumers control 3. Dynamic – add new data at anytime And, much more … 

9 www.xyggy.com © 2019 Thingy