Semi Supervised Learning (SSL)

Hello peeps, let’s get into another interesting topic called semi-supervised learning. According to the data we choose to train there are three types of learning.

  1. Supervised Learning – the traditional learn problems and solve new ones based on the same model again under the supervision of a mentor.
  2. Unsupervised Learning – some lessons in life
  3. Semi-supervised learning – solving some problems on someone’s supervision and figuring other problems on your own.

It all burns down to one simple thing- Why semi-supervised learning and how is it helpful. Without any further ado let’s get started.

Labelled and unlabelled data?

Labelled data is the data which comes under a class and has some properties and unlabelled is what the machine should figure out. Now when we use these both together the machine has interesting behaviour and it starts to perform well. The small proportion of labelled data is clustered and tested on the larger unlabeled population and the classes and their properties are figured out. This is also called as inductive learning.

Assumptions

This process is expensive like any other machine learning algorithm. The three main assumptions in this field of study are

  • Continuity Assumption: The algorithm checks if the points, closer to each other are more likely to have the same output label or less likely.
  • Cluster Assumption: Here we divided data into discrete clusters based on properties and classes and points in the same cluster share the same output label.
  • Manifold Assumption: Here the algorithm checks manifolds and test data of much lower dimension than the input space. This uses distances and densities which are defined on a manifold.

There are different models which are related to semi-supervised learning

  1. Generative models
  2. Heuristic approach
  3. Graph based models

Please read more at Wiki

Some real world scenarios of semi supervised learning

  • One of the implementations of SSL is in AI.  An AI which can identify types of genre movie watched by an audience which of course will require some labelling by an engineer.
  • Climate changes and behaviour in an area – Identifying what type of climate can influence an area’s plantation and the population over there.
  • Movie scripting – there’s a robot that analyzed thousands of movie scripts and scripted its own movie called Sunspring. You can search for “sunspring” on the internet to learn more.

Simply put SSL can do wonders in any field. By implementing SSL we can create AI which can think like human and also performs better than a human.

 

Leave a Reply

Your email address will not be published. Required fields are marked *