How Deep The Deep Learning Is? - Measuring The Depth of Deep Learning

Is it an evolution of Machine Learning?
or
A brand new concept?


Machine is learning   >>  Machine is learning in depth


So, what does this "depth" mean?
- "Depth" comes in the form of hidden layers. The more layer it uses, the more deeper and more complex learning will be.


Is there any flaw or loophole present in Machine Learning that influenced in the birth of Deep Learning?


Well, in Machine Learning, we have to manually select the relevant features of object. The model then references those features for analyzing and classifying new objects.
but
In Deep Learning, the feature extraction process is fully automated. In short, it doesn't need any human intervention. All you need to do is, feed a lot of relevant data into algorithm!


So, how does it even extract features without human intervention?
- Well, it uses Convolutional Neural Network for feature extraction.

Convolutional Neural Network uses hundreds of hidden layers in it's architecture. Every hidden layer increases the complexity of the learned features. e.g. the initial hidden layer learns how to detect edges (can be considered as a basic features), and the last hidden layer learns how to detect more complex shapes of the object we are trying to recognize.

Since in Deep Learning, models are capable of selecting the right features on their own, it resembles to be more natural learning.


Constraints:
- It needs a huge amount of data. The performance improves as the size of the data increases.
- Since it deals with the huge amount of data, it needs comparably a high-performance infrastructure.
- Automated feature extracting process is difficult and expensive in terms of time and expertise.
- Because of its complexity; interpreting and communicating model parameters importance could become vague.


Although a lot of constraints are present in Deep Learning, but when it comes to a field where human written programs don't perform well such as in complex problems like image classification, natural language processing, and speech recognition; Deep Learning can come to rescue in its own deep way!


Do you feel like using Deep Learning for your application? In what kind of applications you would consider using Deep Learning?

Let me know your answer in the comments below and please do share the post with your friends as well!