http://cds.cern.ch/record/979717/files/cer-002642933.pdf

Article Deep learning for supervised classification

Classification

A very easy example for classification, how do email servers know which emails should go into spam folder?

Yep. It classifies the emails into spam and non-spam.

If you notice your spam folder, you should find out that most of the emails of the spam folder are actually spams. Deep Learning is into play here.

But there may occur some exceptional cases. For example, all the emails from TopCoder are sent to spam in my Gmail. You will have to teach Gmail that this particular email is not spam by marking it as important.

This classification problem is actually a very classical one. Another example would be classifying patients based on the tumor they have: benign or malignant.

In fact, in case of medical diagnosis, such applications of Machine or Deep Learning to classify patients based on the types of the diseases they have, has become of great interest among the researchers. And the results are just excellent!

Deep Learning Cuts Error Rate for Breast Cancer Diagnosis | NVIDIA Blog

Clustering

You have a bunch of data but you don’t know anything about their classes. How to understand the behavior of the data you have?

A very intuitive approach is to group kind-of-similar data together to understand their classes that you are not currently aware of, and find out internal patterns and characteristics. You know that this is called clustering.

Where are its applications? A classical example would be market segmentation where clustering is used in variety of ways.

Market segmentation refers to the idea of segmenting or dividing the consumers of the market into some groups. In a group, consumers will be similar to each other based on some predefined set of characteristics. If two customers are not similar based on these characteristics, they are in different groups.

The term predefined set of characteristics may sound vague but it is not. This is actually the properties or features that you know about the consumers, but you did not know any way of classifying the customers based on those properties, until you apply clustering.

Companies use this clustered data and the features of the customers to decide their market strategies, like which group of customer they should target or which group of customers needs more advertising etc. etc.

Another trivial example would be analyzing social networks. There are millions of people in social networking websites and analyzing their behaviors sounds really fun. Clustering plays the role here. This idea of social networks analysis can be extended to real life social scenarios.

Lastly, search engines and many other websites use clustering to group similar web pages, videos, songs etc. and improve results for their users.

When we enter into the World Wide Web, lots of ML things are actually going on all the way.

There are plenty of examples of deep learning methods for classification problems. Any course or book on ML/DL will have that. The most important recent examples probably would be Imagenet classification problems with convolutional neural networks (CNN).

Regarding clustering problems, I would suggest the following papers:

Unsupervised deep embedding for clustering analysis

http://www.jmlr.org/proceedings/papers/v48/xieb16.pdf

Conference Paper Deep Clustering with Convolutional Autoencoders