In 2014, Google paid over $ 400 million for Deep Mind Technologies – a London-based startup focused on deep learning research. The company was working on a face recognition technology based on the video, as well as the text extraction from the audio file.
What is so attractive and profitable in the deep technology that the tech giant was buying startups for so much money? Deep learning is one of the fastest growing branches of artificial intelligence. How does it work? Simply speaking, in deep learning scientists try to improve technology to create neural networks, in other words, they are trying to develop IT systems that will be constructed and work as a human brain does.
The neural network is an enormous amount of interconnected processors running at the same time. Each of them has access to local storage and is powered by a significant amount of data. Each one of them also obtains information on the relationships between the data. If we want the neural network to “learn” well, the program must show it how to behave in response to specific external stimuli (i.e., entering data by a computer user).
Deep learning technology, or rather the neural network, has existed for fifty years, but at the end of the nineties, it was not that popular due to the lack of sufficient data. So, what made the situation change in less than two decades? First of all, we have a lot more data needed to build a multi-layer neural network. Secondly, with the latest technologies, we now have machines with much more computing power.
The human brain contains an average of about one hundred billion neurons, and each one is connected to another ten thousand neurons, so the number of connections between neurons – synapses – ranges from 100 billion to 1,000 billion. That’s a lot… That’s why we haven’t succeeded in building a neural network of this size and with such high computing power yet (though Google is working on the creation of artificial neural networks comparable to the brains of laboratory mice).
In order to make the best imitation of the human brain, researchers create complex neural networks spread over many layers. In fact, that is the multilevelness creates a possibility of deep learning. It is worth noting that deep learning is more like teaching a little child rather than a traditional programming because it is based on the cognitive processing. This allows machines to “understand” human signals and “respond” in a way that is understandable to humans.
Deep teaching is based on the mechanism of human brain work – network connections create tangled layers on many levels. Teaching the system involves “updating” the calls after each new stimulus. A machine based on deep learning implies the behavior attributed to humans – to imitate voice, identify images, or predict.
Deep Teaching Vs. Machine learning
The articles on artificial intelligence and the fact that we should seriously consider its potential can be found all over the network. Each of us has probably already bumped on some information on both machine and deep learning. The nuances between these three technologies are apparent to experts. What about the rest? What are the most significant differences between machine learning and deep learning and can one say that one has the upper hand against the other one?
Let’s start with the weakness of machine learning:
- It requires the presence of a human who, by introducing thousands of examples, teaches the machine how to draw knowledge from them
- All errors must be corrected manually
- The learning process itself is very time-consuming
- The machine is fully human dependent, so it is difficult to determine its level of intelligence
Deep learning takes place without supervision and control. I write “rather” because there are known cases where deep learning requires a human presence. However, for the most part, deep learning is the creation of neural networks that allow the computer to think independently (without human participation).
The power of the deep learning model comes from the data. It needs more than ten thousand data cases to achieve good results. Of course, this number is not the only requirement, because the data must include many variations so the machine could understand it.
Artificial Intelligence is like an umbrella that covers both deep learning and machine learning. If we were to draw a circular diagram illustrating the relationship between Artificial Intelligence, Machine Learning, and Deep Learning, then the largest circle would be reserved for artificial intelligence, in the middle, there would be a “field” for machine learning, and in its very center – for deep learning.
Deep learning in e-commerce
Maybe most people still think that this technology is still commonly unknown. Well, no. Google’s translator is using deep learning technology. Moreover – two years ago Microsoft created a deep learning algorithm that is more accurate than a human. During the test, it classified faultlessly over three hundred thousand photographs into one of the three descriptive categories.
So, how deep learning can be used in the e-commerce? First and foremost, it will be much easier to get even more accurate demographic data, because computers are constantly being “taught” to differentiate between age and other indicators. Also, traffic analysis in the store can be more detailed, because we have more knowledge of the activities that the potential buyer do in the shop, who is he, and what actions he is undertaking on the site.