Deep Learning is one of the most talked about topics in the present time. The rising trend of this field of study can be easily visualised by the use of this term in the search queries.
It is interesting how the use of this term goes on increasing day by day. Deep Learning has steadily increased in growth as a field of interest. Its ideas and its achievements are all around us. It is responsible for most of the new innovations that we hear of. It is credited for the recent improvements in the fields of computer vision, language translation, natural language processing, driver assistance systems, autonomous driving, medical diagnostics etc.
So, what is deep learning and how is it able to accomplish all these feats and what is it that has to lead to this outbreak of deep learning techniques in this decade, we will see all about these questions and I will try to answer most of them. By the end of this article even though you might not become an expert in deep learning but you might be equipped with most of the ideas to get you started in this field or even have a fruitful conversation with anyone interested in this field.
Since this is a quest to quench most of the doubts we have about deep learning, we will enlist the most prominent of the doubts and go on finding an answer to these questions one at a time. The questions are:
- What is Deep Learning?
- Why is Deep Learning used?
- How is it used?
So let’s get started with our first question.
What is Deep Learning?
According to Wikipedia, Deep Learning belongs to the broader family of Machine Learning methods that are based on learning data representations as opposed to task-specific algorithms. Even though this statement sounds a bit condensed it encompasses many interesting ideas. So let us analyse this statement to get a clearer picture.
Deep Learning is a small subset of a pool of methods called Machine Learning. The methods in machine learning employ a different way of problem-solving than the traditional task-specific algorithms. In task-specific algorithms, an algorithm is made to solve a specific task by making use of the domain-specific knowledge of the creator. Such algorithms usually fail to generalise if the problem at hand changes even by a little margin. These algorithms are also at a disadvantage because the programmer is supposed to anticipate all kinds of situations that the algorithms will face.
But instead, in a machine learning algorithm, the system is allowed to automatically discover the patterns needed for feature detection or classification from raw data. This replaces manual feature selection and allows a machine to learn the features and use them to perform a specific task.
In other words, given a set of correct examples, the algorithm will ‘understand’ the patterns to ‘learn’ how the examples are correct and from the ‘knowledge’ gained it will be able to solve the questions given to it to produce correct answers. The words ‘learn’ and ‘understand‘ in this context have a much deeper meaning and we will be talking about it later.
Since we now have got a picture of what the machine learning algorithms are, let us come back to our point of interest i.e. ‘Deep Learning’. If all the machine learning algorithms learn patterns, what are the criteria that differentiate them?
It is the way in which the patterns are learnt differentiates these machine learning algorithms. Deep Neural Networks or Neural Networks, in general, are biologically inspired. They try to mimic the learning process of the human brain. Different neural networks try different types of learning methods of the brain known to us.
Convolutional Neural Networks (CNNs) is such an example where its learning process inspired by the Visual Cortex of the brain.