How to Build a Learning Machine
What is going on here?!
Google’s image recognition algorithm is analyzing your lines in real-time and providing feedback. By drawing, you are adding material to the neural net, making the algorithm better at recognizing the drawings it encounters.
Here is how they explain it:
What separates an umbrella from a mushroom?
If you draw things for a living, it’s pretty important that you have a nuanced understanding of why we see what we see. The algorithm thought that my umbrella could also be a mushroom. But several features of my drawing gave the neural network high confidence in interpreting my lines as “umbrella” rather than “mushroom”. I think we can become better visual communicators by dissecting the key features of the objects we draw. I think of an umbrella as a collection of three particular features:
Look through these images again, analyzing how people have drawn the handle of the umbrella. If there is a “J” indicating a handle, I see an umbrella. If it is a straight line instead, I see a parasol, or large standing umbrella.
Machine Learning or Learning Machine?
From the point of view of Google’s software engineers, you the artist, are training their algorithm. This is machine learning. But we can turn the tables! As artists, we can use the algorithm as a learning machine.
Here are the steps:
- Go to quickdraw.withgoogle.com.
Use the suggestions to practice drawing from imagination.
- Focus on simplicity and clarity.
If Quick, Draw! shows you the word “umbrella”, aim to draw the simplest, clearest image of an umbrella that you can think of.
- Review examples drawn by other people.
When you’ve completed six drawings, click on each to see how other people have drawn the same objects. What can you learn to improve the clarity of your drawings?
The practice of drawing and comparing your doodles with those in the algorithm’s library will train you to get better and better and communicating clearly through drawings. Give it a go now!
This was a happy and welcome email this morning. Thank you Dorian
Cheers Michael 🙂
This would be a really great way to train for Pictionary.
Fascinating stuff. Thanks, Dorian, for sharing.
Super idea and Great fun! Thanks Dorian
I enjoy reading your articles Dorian, you’re doing great man. 🙂
au ich freu mi über dini publikatione! einigi hani uusprobiert anderi uf nach “furchtlos” verschobe. dänn wotti meh lerne.
das da obe isch merk-würdig und luschtig!
teile macht wach. danke Sohn
Das freut mich! 🙂