Evolution of Artificial Neural Networks

Have you figured out Google and Facebook got smarter than it was few months back. When you type something in the Google search, did you notice that it will instantly predict what you are going to search? It may not just based on the words, but it is kind of knowing what your behavior of last couple of days. For an example, if you are looking for a frock which you are willing to wear for a dinner outing on coming week, in your favorite color, you will see similar dresses everywhere on your social media. You will see a lot of advertisements of dresses in same nature In your news feed. Soon you may have figured out that the intention you had in your mind is visible to rest of the world. It happens not only for you, but to billions of internet users daily.

Facebook AI

If we wrap out the mechanism in those multi billion companies which observes your behavior, you will find out a common structure which has been built very recently in their core network replacing old algorithms like page ranking . It is called deep neural net which has the capability if mimicking the human brain.

The idea was formed back in 1950s from one of the revolutionized paper published by Allen Turing called “Computing machinery and Intelligence” which based on the Turing Test. After then scientists were working on this interesting topic on building a machine which can think with the military funding from US and UK governments particularly. It was an exploration on neurology, mathematics and computing. Neurologist have found out that the brain works with electrical signals which are firing and upon a thought process. Norbert Wiener’s cybernetics described control and stability in electrical network. And Alan Turing’s theory of computation showed that any form of computation could be described digitally using ones and zeros. Based on these 3 pillars, Artificial intelligence grew up.

In 1970s, the enthusiasm of artificial intelligence has lost due to lack of computational power. And most of the mathematicians criticized the idea because the basic principles of AI did not contain a strong mathematical backup. And scientists could not convince their results because of the limited computational power so that most of the funding were stopped. This era is known as the winter in Artificial intelligence.

As the moor law predicts, computational power has doubled every two years so that again scientists could work on the theories they have formed in the winter. It worked. The works on AI is boomed in the first decades of the 21st century.

Human Evolution

Enough about the history, what is this so beauty of Neural Networks and artificial intelligence? In order to answer this question let me briefly describe what an artificial neural networks do in simple terms. It is a system which can predict something based on the past experiences. It is like a child get trained to stand up on his own. After so many attempts, the child knows now to keep the balance when he is standing. How to keep his body straight, what is the distance he should maintain between feet, what is the pressure he has to put on each part of the limbs and so on. Like that this computer programmed network can learn any task adjusting the weights of its metrics(tensors) at each try. After it is been fully leaned, it can generate an output closer to the desired output for a given input with a calculated accuracy and learning rate.

Importance of Artificial Neural Nets can be matched with the ‘human evolution’ proposed by Da-vinci. Our body, our brain, our limbs has developed, evolved throughout million years going via uncountable number of iterations,passes and misses. Not only for human beings, everything on the earth has evolved through million years and produced the optimum shape and form for the current existence. From a living cell, an unimaginable number of iterations might have gone through and our DNA has corrected its biological memory at each pass. Similar to that we run an artificial neural network thousands or millions of iterations so that its weights can adjust for out trained task.

Currently most of the developers are working on image recognition and speech recognition and its application since it has a wide range of applications such as character recognition, object detection, abnormality detection and etc. And it is about to hit the Robotics arena as well. Where robot manipulations are done using neural networks. It is assured that coming years will be the summer in artificial neural networks and for more development. People will use this technology and there might be a decade where we will question the growth of artificial intelligence over humanity. Because it took millions of years to learn certain task optimally for human. Now we can shrink it off to few hours of learning time.And Yes, It is scary!!.

Like what you read? Give Kasun Vimukthi Jayalath a round of applause.

From a quick cheer to a standing ovation, clap to show how much you enjoyed this story.