Deep Learning: hype and value in the industry
Deep learning is the “new trend” in machine learning, but… what is it exactly? Why is it receiving growing interest? Which are its practical applications? Finding good answers to these questions is not so easy. In this article, we are going to introduce some basic concepts about deep learning and take a look at the potential of a technology that could really revolutionize our lives.
What does “deep learning” mean?
The term deep learning refers to a set of machine learning algorithms relying on multiple processing layers to model high-level abstractions in the data. For our purposes, deep learning is simply a synonym of multiple hidden layers artificial neural networks (ANN) (have a look at A basic introduction to neural networks for more information).
Unlike “shallow” ANNs (single processing layer ANNs), deep neural networks seem to better model the cognitive process: each new layer aggregates the information/features extracted by the previous layers to build something more complex; it is a hierarchical process that tries to emulate the way humans perceive the real world.
That’s why deep learning is particularly suitable for classification tasks, image recognition and generation, speech recognition, object detection, recommendation systems, predictions and much more; that’s why deep learning is overcoming a lot of AI problems that seemed to be impossible to solve until a few years ago: no one would ever have guessed that a program could beat the Go world champion (Go is considered one of the most difficult games ever) who after the defeat stated: “If no one told me, […] I would think the player was a very strong player, a real person“. We are talking about software that could emulate the thinking process; deep learning might effectively reduce the distance between humans and machines!
Why now?
You are probably wondering why a so incredibly promising trend is having exponential improvements only in the last decade and that’s a right question, considering that ANNs have existed for almost 50 years! However, there are at least two things we have now that were not previously available:
- Big Data: a huge amount of data we can use to train our ANNs
- Highest computational power: cloud computing and graphics card processing to train increasingly deeper systems
Deep learning as a means to improve our everyday life
Thanks to deep learning a growing number of companies gives real value to data and extracts useful hidden information that could be used to make better decisions and improve their revenues: we are talking about data-driven companies.
And there is actually more:
- Deep learning might be used to improve the client satisfaction and the quality of services offered to the client, in general. By predicting server failures, for example, companies could minimize the downtime, reduce money losses and would have more resources to improve their services.
- It could have medical applications (allowing to accurately predict diseases and/or building intelligent robots that help doctors to perform minimally invasive surgery).
- It might improve our “everyday security” (just think of self-driving cars, video surveillance, military applications, …).
We do not know what tomorrow will bring, but it would not be wrong to say that deep learning is marking the beginning of a new era.
If you are interested in this topic and you would like to know more about deep learning, leave a comment and keep following!
Graziano Mita