Machine Learning – Introduction

Why Learn Machine Learning?

Machine Learning (ML) is currently one of the most talked about topics in the field of technology, and will remain so for the coming years. The reasons for this are:

1. The high amount of data being produced today by us in the form of text, images, videos, geolocations and many more.
2. Increase in computation power, thus leading to higher speed and lower cost of analysis of the data.
3. The development of better algorithms, leading to better analysis, better prediction power and higher accuracy.

Today Machine Learning is being used everywhere, from mundane to high end applications.

A few examples of mundane applications of Machine Learning are:

• Google search and page ranking
• Recommendation engines for youtube videos, amazon shopping etc.
• Face detection in image tagging
• Language translation (using Google translate)

A few examples of high end applications of Machine Learning are:

• Cancer detection from x-ray images
• Self driving cars for automatic navigation
• Aircraft Control for altitude adjustment
• Computer vision in manufacturing, such as detecting defective samples
• Computer vision in security systems for threat detection

Machine Learning is a continuously developing field that will keep covering more and more areas of applications in the coming years and will also keep improving its performance on all of those areas, thus driving technological progress and business growth.

What is Machine Learning?

• Machine Learning is a sub-field of Artificial intelligence. Machine Learning itself can have other sub-fields such as Deep Learning which is a class of machine learning algorithms that are inspired by human brain.
• Although AI and hence machine learning is a field within computer science, it differs from traditional computational approaches:
  • In traditional computing, algorithms are step-wise instructions that are programmed explicitly by us and used by computers to solve a problem.
  • On the other hand, Machine Learning algorithms learn from the data provided to it, using statistical analysis/modeling approach and make their own set of rules. The ML algorithm can then make decisions based on these set of rules that it learnt from the data.

Types of Machine Learning

Machine learning algorithms can be classified into 3 broad categories:

1. Supervised Learning
2. Unsupervised learning
3. Reinforcement Learning

ML algorithms are classified into these three groups based on the type of data used for learning, and the way feedback of each thing learnt from the data is provided back to the model.

1. Supervised Learning
   • In Supervised learning, the ML model is presented with training data which is labeled, which means that each data is tagged with correct labeling.

     

     The ML model goes through this labeled data and learns the pattern/rules from this training data. It learns by creating a function mapping from input data/variables to output data. This allows the model to becomes capable enough to predict new unseen data, also called ‘test data’.

   • Supervised learning  can be further divided as:
     • Classification: This is used when the output variable (variable to be predicted) is a category.
       For ex: predicting from a given input data whether a particular email is spam or not spam; predicting from certain environmental conditions (input variables) whether the weather will be sunny, cloudy or rainy; etc.
     • Regression: This is used when the output variable are numerical values ex: weight of students in a class
2. Unsupervised Learning
   • In unsupervised learning, an ML model is given a set of training data that is not labeled with correct output values (unlike supervised learning). Here, the model finds the pattern in the data not by learning from true labels (Output variables) but rather by inherent pattern of the input data itself.

     

      

      

   • Unsupervised learning  can be further divided as:
     • Clustering: Discovers the inherent pattern within the data and forms groups, such as grouping customers by their purchasing behavior on an e-commerce website, such as amazon.com.
     • Association: These kinds of algorithms learn association rules from the data. These association rules describe a large portion of the data, for ex: people who buy a glove are also more likely to buy a ball.
3. Reinforcement Learning
   • A reinforcement learning algorithm, also called an ‘agent’, learns by interacting with its environment. When the agent performs an activity correctly/accurately, it is given a reward, and when the agent performs the activity incorrectly, it is penalized.
   • By maximizing the ‘reward’ and minimizing the ‘penalty’, the agent learns a particular task without any intervention from humans.
     • Thus reinforcement learning is a type of dynamic programming that trains a machine learning model using a system of reward and punishment/penalty.
   • Example of Reinforcement Learning:
     As shown in the image, the agent is given two options, a path with water or a path with fire. If the agent uses the fire path then it gets a penalty by subtracting 50 points from it, in this way the agent learns that it should avoid the fire path.If it chooses the water path which is the safe path, then 50 points would be rewarded to the  agent. In this way the agent learns that water path is the safe path.In this way, using a reward-penalty system, the agent would learn, which path is safe and which isn’t. Thus improving its environmental knowledge to select the next action.