I had a play around with Google Quickdraw, an online doodling game/neural network created by google that collects data sets of doodles. Using the worlds largest dataset of doodles to train the neural network to guess drawings in under 20 seconds. It currently consists of over 50 million drawings contributed by over 15 million people, also having the advantages of being open-source and being developed by Google has the standing in our society to collect the massive sample size.
As shown it had no problem recognising my very quick mouse-drawn rabbit along with nearly everything else I was asked to draw after 3 rounds of the game. The only one it couldn’t guess was ‘a short monster’ which I thought was quite interesting and propelled me into looking deeper into what defines a ‘monster’. The dictionary definition of a monster is “a large, ugly and frightening imaginary creature”, I feel as though people have there own personal various views of ugly and frightening, as well as varying imagination. The fact that monsters are imaginary and products of peoples imagination that we see in films etc. means there is endless ideas of what a monster may look like, making this specific category more difficult but not impossible for the neural network.
There was a lot of variation in the doodles it learned from for this specific category, most others for example pizza only had 2 general variations that people drew(either a slice or a whole pizza).