Will scientists ever be able to make a machine that mimics the human brain? So far, scientists have been able to create computers that can recognize patterns, learn, and make accurate predictions. These artificial neural networks use a combination of hardware, software, and mathematics to work on a variety of problems.
Machines That Learn
Neural networks consist of a group of nodes. Each node performs a mathematical operation. Some of the most complex neural networks use polynomials to simulate the way our brains make decisions. When an engineer constructs a neural network to solve a problem, he has to train it. The network receives data and makes predictions based on that data. During the training phase, it receives feedback on whether its answers are wrong or right. It rearranges itself until it can consistently achieve correct answers, essentially “learning” from its mistakes. Scientists have trained neural networks to determine whether tumors are cancerous. Other neural networks help recognize faces, speech, or handwriting. Google recently pioneered one of the world’s largest neural networks and then taught it to recognize videos of cats.
As of now, artificial neural networks can only tackle a single problem at a time. Regardless, they are extremely helpful to scientists, capable of finding patterns and connections that their human counterparts might otherwise miss. As time passes and neural networks become increasingly complicated, researchers hope to create robots with neural networks as complex as the human brain’s (such as the one partially mapped in the image above).
Check out the links below to learn about neural networks that diagnose cancer, play poker, and simulate the brain signals that control walking.