Programs like deepart.io are so fascinating. You give the program a photograph and an artwork, and then it transforms the photograph to look like the style of the art.
A photo from my parents wedding combined with Juan Gris’ “Portrait of Picasso” generates a cool result.
That’s a deep learning algorithm called neural style transfer.
On New Years Eve, I was reading about Machine Learning. You can get two books on Machine Learning and Python for $1 (via humble). In these books I was learning how you feed thousands of images of a subject matter to a program, and it learns to identify and create new images of that subject matter.
For instance, if you give thousands of images of apples to this program, it will learn how create its own images of apples. These created images are completely artificial intelligence created.
This learning of what subject matters are made me think, since I have a six-month-old at home. We are always teaching her things, pointing out things, saying the name. Whenever we do that, it’s always an isolated example. Here’s a drawing of an apple. “This is an apple.”
But the best thing for her to learn would be to get 500 images of an apple. Show her the 500 different images of an apple. Just like how you would do with Machine Learning.
There should be kids books where each spread has 50 different images/interpretations of one thing.
If I saw 50 images of an apple, that would help me to learn what an apple is—much better than an image of one apple. These would be various types of apples. Mot just the same apple. It would be photos of apples in different contexts and drawings of apples in different styles.
Most kids books are illustrations. rarely ever photos. I wish kids books had more photos.
Seeing lots of various drawings and photos of apples would show a child what is common among apples. I’m going to make some books for my daughter. Maybe sell the books on Amazon.