I started getting myself familiarized with machine learning in the last couple of months. In this series of blog posts I will try to organize and summarize the things I learnt. Subsequent posts will also present final outcomes of few exciting projects I have completed. Let’s get started.

What is Machine learning?

Machine learning is a branch of artificial intelligence that is based on the notion that systems can learn from data, identify patterns and make predictions/decisions with minimal human intervention.

Machine learning classification

There are four basic types:

1. Supervised learning
2. Unsupervised learning
3. Semi-supervised learning
4. Reinforcement learning

The final choice of the approach depends on what type of data you are dealing with. In general these approaches contain three core components: representation, evaluation, and optimization (most relevant for supervised learning). Let’s understand what each of these approaches mean.

1. Supervised learning

In this method, a data scientist supplies an algorithm with labeled training data, define variables to assess for correlations and matching output values. In other words, a data scientist teach a machine using examples. While learning, the machine identifies patterns in data and then makes predictions. As you can see, the model building portion involves significant work and hence called supervised learning. This is the most frequently used form of machine learning and is great for regression modeling (to predict continuous values), binary or multi-class classification (to divide data into two classes or choose between >2 answers) and ensembling (to combine multiple ML predictions to generate an accurate prediction).

Supervised learning methods

There are many complex supervised learning algorithms. However, they all are broadly divided into two main classes:

1. Classification: These models are used when the output variable can be categorized (i.e., Yes/No, Pass/Fail, benign/malignant/normal, etc.).
2. Regression: These models are used when the output variable is a real value, hence can be used to predict or forecast a numeric value based on previous observations.

The following table shows most frequently used algorithms in each of these classes:

2. Unsupervised learning

As the name implies, this mode of machine learning does not require data to be pre-labeled. Algorithms in this class can sift through data to identify patterns that then can be used to categorize data into different classes. These learning methods are great choices for clustering (to split data into groups based on similarity), anomaly detection (to identify odd data points in a data set), association mining (to identify sets of items in a data set that frequently occur together), and dimension reduction (to reduce the number of variables in a data set).

Unsupervised learning methods

There are many complex unsupervised learning algorithms and they can be broadly divided into few classes as below:

1. Clustering: This is a data mining technique which groups unlabeled data into groups based on how similar or different they are to one another.
2. Association rules: An association rule is a rule-based method for identifying relationships between variables in a given data set. Have you ever wondered how businesses decide a pair of products for an offer like “buy one get one 50% off”? This is a great example of association rule learning.
3. Dimensionality reduction: This is a technique used when the number of features, or dimensions, in a given data set is too high. It reduces the number of data inputs to a manageable size without compromising the integrity of the data set.

Following table shows most frequently used algorithms in each of these classes:

3. Semi-supervised learning

This machine learning method combines features of supervised and unsupervised learning. In this method, the model is built using both labelled and unlabelled data. In many applications, annotating data is cumbersome. In this method, the algorithm first use labelled data to understand dimensions and accordingly learn to label unlabelled data. As you may imagine, semi-supervised learning strikes a middle ground between the performance of supervised learning and the efficiency of unsupervised learning. Few areas that benefits from these methods include machine translation, fraud detection and labelling data.

4. Reinforcement learning

This learning method is often used for robotics, gaming and navigation. With reinforcement learning, the algorithm discovers through trial and error which actions yield the greatest rewards. There are three primary components of reinforcement learning: the agent (the learner or decision maker), the environment (everything the agent interacts with) and actions (what the agent can do). The objective is for the agent to choose actions that maximize the expected reward over a given amount of time. In other words, the agent seeks positive rewards, which it receives when it performs an action that is beneficial toward the ultimate goal and avoids punishments, which it receives when it performs an action that is gets away from the ultimate goal. The agent will reach the goal much faster by following a good policy. So the goal in reinforcement learning is to learn the best policy.

This concludes the first machine learning tutorial. In the next tutorial let’s dive into learn the steps involved in machine learning. Please leave a comment below if you have any questions/comments.