Artificial Intelligence, that is creating machines that can understand logic and perform tasks by themselves as if they were independent thinking organisms, started its journey in the nineteen forties and fifties. Most of the early researchers of artificial intelligence believed that the key to AI is logic. Their focus was on teaching the computer to interpret logic by providing it with examples to follow. However, there were two scientists who did not like this preoccupation with logic. They had a different idea, something unbelievable, and something crazy.
They were Alan Turing and John Von Neumann. Their crazy idea was to adopt the working principles of the human brains, its process of cognition and schema formation to create artificial intelligence.
As we all know, they did not live long enough to prove themselves right but here we are in the second decade of the twenty first century reaping the benefits of deep learning – that crazy approach.
Deep learning is the epoch of machine learning
So far as machine learning has developed deep learning is its most advanced exponent. A simple illustration can explain this. Suppose you need to create a machine that can recognize a cat. If you are using traditional machine learning you would have to extract certain features of a cat and create algorithms that can look for those features. The problem with this approach is that it is too dependent on the human brain. The human has to figure out what features to look for and he or she might miss something.
Deep learning on the other hand uses layers of networks which resemble the neural networks of the human brain in functionality. It completely skips the feature extraction part and recognizes all certain nuances of the cat’s being.
The accuracy achievable in the second method is certainly higher.
How deep learning works in the real world
There has been some drastic improvement in the field of computer vision. The image processing technology from the old world has been replaced by image recognition with the help of deep learning. Almost all state-of-the-art object recognition programmes use deep learning the most familiar example being Google photos. You can search for a certain image without it having been labelled earlier.
The field of speech recognition and language understanding has also taken a leap with the help of deep learning. The neural network can create a language model from a textual data base. Initially it tries to interpret the sequence of characters in the text, makes random guesses about the characters that may follow each other. But gradually it understands the vocabulary and before you know it can write whole chapters for you.
Deep learning training at this point in time can really take you places; the prospects are unimaginable. The learning curve is steep so it is better to get at it already.