What is Deep Learning?

 

What is Deep Learning?

Deep Learning is actually subfield of Artificial Intelligence and Machine Learning that is inspired by human brain. To understand this have a look at the following diagram:

Machine Learning learn from data that contains input and output but for this relationship, ML algorithm depends on statistical techniques whereas Deep Learning depends on logical structure called neural network inspired by human brain. 

So we can define Deep Learning as:

Deep Learning is a part of a broader family of machine learning methods based on artificial neural networks with representation learning. Deep Learning algorithms uses multiple layers to progressively extract higher-level features from the raw input.

Why the name "Deep Learning"?

Because of lots of hidden layers being used here, the network becomes deep, hence the name deep learning. 

What is Representation Learning?

Now the question is what is actually representation learning? Deep Learning, unlike Machine Learning, extracts features automatically. We don't need to do feature engineering like we need to do in ML. This is actually called Representation Learning. Initial Layers detect simple features and deeper layers detect complex features.

Why is DL getting so famous?

There are two most important reasons:

1) Applicability:

Deep Learning algorithms are algorithms are applicable in lots of fields.

2) Performance:

Its performance is so good that it can even beat human intelligence. 

DL VS ML:

1) Data Dependency:

Deep learning needs more data as compared to ML. You can say that DL is data hungry. ML models work on less data but after certain amount increase in data won't do any good to performance.

2) Hardware Dependency:

ML models can be trained on CPU but there are complex matrix multiplication in DL hence you need GPU. On CPU it will train slow. You need more powerful GPUs with more memory hence costly.

3) Training Time:

DL models are complex and hence training time is high as compared to ML. It can take weeks or even months to train a DL model but ML models get trained in minutes and hours. Training Time is slow in DL while Prediction Time is fast.

4) Feature Selection:

DL models extract features themselves, only data should be in right form. In ML you have to extract them manually.

5) Interpretability:

DL has interpretability issue. If it did a prediction, we don't know why it did that. For example, if it predicted a picture as dog we don't know why it predicted it as dog and not cat. DL is just like a black box. If you need to give reason for certain prediction then DL is not a good choice.

Hence, you can't replace ML in every project. You can use DL and ML based on situation.

Conclusion:

This was all about today's post. Hope you liked it. If you have any queries just comment below!