I got lost in an absolute myriad of thoughts the other day, and it essentially wound up wondering if we can teach machines to count, beyond of what it can see in an image, and I've come up with a small experiment that I would absolutely love to collaborate on if anyone (@Google?) else is interested.
The idea is based on the concept of the experiments performed using "Wisdom of the Crowd", commonly in this experiment to use a jar of jelly beans and asking many people to make a guess as to how many is in there. Machine learning can be used to make predictions from patterns, but it would have nothing to gain looking at one picture of a jelly bean jar to the next and being able to correctly identify that is in fact - a jar of jelly beans.
But suppose we feed it several images of jars of jelly beans, along with all of the guesses people have made of how many is in there. Can we then presume that feeding it a new image, it would be able to give us a fairly accurate count of how many there are by itself?
Let me know if there are any interested parties willing to take up this challenge!
The idea is based on the concept of the experiments performed using "Wisdom of the Crowd", commonly in this experiment to use a jar of jelly beans and asking many people to make a guess as to how many is in there. Machine learning can be used to make predictions from patterns, but it would have nothing to gain looking at one picture of a jelly bean jar to the next and being able to correctly identify that is in fact - a jar of jelly beans.
But suppose we feed it several images of jars of jelly beans, along with all of the guesses people have made of how many is in there. Can we then presume that feeding it a new image, it would be able to give us a fairly accurate count of how many there are by itself?
Let me know if there are any interested parties willing to take up this challenge!
Comments
Post a Comment