Scientists Model Brain to Teach Computers to RecognizeResearchers at the beginning of the computer revolution assumed that teaching a computer to recognize something would be easy. But, half a century on, computer vision is still a primitive thing. One group of researchers have made a stride towards fixing that, however.
Researchers from Los Alamos National Laboratory, Chatham University, and Emory University created a neural network with a slightly different structure than what is usually used in research. You can think of neural networks as being separated into layers. Most neural network research involves wiring artificial neurons from one layer to others in another layer. This gives the network some ability to recognize patterns, but they haven’t been truly successful with vision, yet.
In this case, the researchers decided to wire some neurons to other neurons in the same layer, creating lateral connections. while it might seem a little counter-intuitive, doing so gave the neural network a greater ability to recognize things than other neural networks.
"Lateral connections have been generally overlooked in similar models designed to solve similar tasks. We demonstrated that our model qualitatively reproduces human performance on the same task, both in terms of time and difficulty. Although this is certainly no guarantee that the human visual system is using lateral interactions in the same way to solve this task, it does open up a new way to approach object detection problems,"said Vadas Gintautas of Chatham University in Pittsburgh.
It turns out that our brains rely heavily on lateral connections when analyzing images. These connections are critical in their ability to parse and pare down information into something other layers can further process. Think of it like a series of filters: at the top level, a general filter is triggered, say, the object is boxy and long. The next layer takes a look at the boxy and long signal and sees that the shadows mean it has depth and is shaped a bit like an L. The next level sees that it is a couch, and then the next sees the details of the couch. Each layer acts as its own sub neural network, communicating information around to find a pattern that matches.