Estimation without Counting Observed in Artificial Neural Network
Researchers have been wondering for a long time how humans learn. After all, no one is born with an understanding of math or language, yet both of these develop throughout early life. In the case of math at least, many forms of life, including humans, have demonstrated an ability to understand when one set is larger than another, without counting the items. Now a virtual neural network has done the same.
This neural network was designed only to mimic the retina of an eye and then generate false images, similar to what it originally saw. How the neurons fire as the original image is viewed and the false ones made is recorded. The researchers found the lowest level of neurons, those furthest from the virtual retina, were firing based on the number of objects in the original image, despite the fact that there is no understanding of numbers in the program. This information was then given to a second program which was able to estimate whether the image had more or fewer objects than some reference number the researchers also gave it.
This finding could be very important for understanding not only how humans learn numbers, but also dyscalculia and robotic vision. Dyscalculia is a condition which makes it almost impossible for a person to acquire even basic math skills.