Machine Learning : An Introduction

This article will help understanding what exactly is machine learning and how is it related to / different from Artificial Intelligence, Data Mining, Deep Learning, Statistics, Recommendation Systems, etc.

What is Machine Learning ?

Machine learning is a set of techniques that uses algorithms which iteratively learn from data and enables machine to make and improve predictions and behavior.

How do we teach a machine ?

Machine Learning Process

1. Collecting Data: Data-set having variety, density and volume of relevant data will help in better learning.
2. Preparing the data: This involves fixing issues with the data set collected e.g. handling outliers and managing missing data points. Break the cleaned data-set into two parts, one for training and other for evaluating the program.
3. Training a model: Choose an appropriate algorithm and representation of data in form of the model suited for your problem. Use the training data-set to train the model.
4. Evaluating the model: To test the accuracy and precision of the model, use the test data-set kept aside in the step 2.
5. Improving the performance: It might involve choosing different model and algorithm altogether, or introducing more variables and/or data to train the model.

When do we say that machine is learning from some data?

“A computer program is said to learn from experience E with respect to some task T and some performance measure P, if its performance on T, as measured by P, improves with experience E.” — Tom Mitchell, Carnegie Mellon University

In layman's’ terms, if a computer program improves itself without explicit enhancing the program/algorithm, instead using the data that it receives generally by some feedback signals. For e.g. A search engine improving its result based on the user clicks.

In the example, the task ‘T’ of the program (search engine) is to provide relevant result, which has the performance measure ‘P’. When the user interacts with the program it receives data (user’s click) which is like the Experience ‘E’. Now if the program analyses the data to boost the results which have higher clicks for a particular search keyword, will will be increasing its performance measure ‘P’ as the users won’t have to scroll down a lot to get to a possible relevant result.

Where do we use it ?

• Search results (by google and others)
• Fraud detection (in transaction by internet/mobile banking)
• Real-time ads (e.g. youtube)
• Recommendation on e-shopping sites (e.g. Amazon — Customer who bought this also bought)
• Pattern and image recognition (e.g. Facebook’s tag suggestions)
• Email spam filtering (e.g Gmail, Outlook, etc)

Similarity / Difference : Machine Learning & …

There are so many fields like Artificial Intelligence, Data Mining, Deep Learning, Statistics, Recommender Systems, Big Data, etc. which are closely related but sometimes misunderstood as same thing.

Machine Learning and Artificial Intelligence

Artificial Intelligence is a technique to enable computers to take rational decision. Machine learning is a subset of Artificial Intelligence where the machine is trained to learn from it’s past experience.

Artificial Intelligent system posses the following (in general):

• Natural Language Processing to enable it to communicate successfully in human language
• Knowledge representation to store information provided before and during the process
• Automated Reasoning to use the stored information to answer questions and to draw new conclusions.
• Machine Learning to adapt to new circumstances and to detect and extrapolate patterns.

Machine Learning and Deep Learning

Deep learning is a specialized type of Machine learning which is associated with Artificial Neural Network (ANN, a machine learning algorithm), that uses the concept of human brain to facilitate the modelling of arbitrary functions.

ANNs can -in theory — model any kind of relationship within a data set, but in practice getting reliable results from neural networks can be very tricky.

Machine Learning and Data Mining

Data Mining deals with searching specific information and Machine learning solely concentrates on performing a given task.

E.g. : Teaching someone how to cook is machine learning and finding the best chef trainer in the city is Data Mining.

In terms of the pattern discovery, following is the major difference:

• Data Mining discovers previously unknown patterns and knowledge
• Machine Learning is used to reproduce known patterns and knowledge, automatically apply that to other data, and then automatically apply those results to decision making and actions.

Machine Learning and Statistics

Statistics is a branch of Mathematics which utilizes data, either of the entire population or a sample drawn from the population to carry out the analysis and present inferences.
Machine learning uses the statistics heavily in its algorithm.

Any statistical modelling assumes a number of distributions while machine learning algorithms are generally agnostic of the distributions of all attributes.

Popular Machine Learning Methods :

Machine learning tasks are typically classified into three broad categories, depending on the nature of the learning “signal” or “feedback” available to a learning system.

Supervised Learning (Predictive Models):

The system is presented with example inputs and the desired output, and the goal is to learn a general rule that maps inputs to outputs. These are used to predict the future outcome based on the historical data.
For a given set (x as input,y as output), the algorithms will try out to find a find a function ‘f’, such that y = f(x)

Assuming there are no other variables, for given (x,y) let the function be in form of
y = f(x) = a + b.x
The task of the algorithm is to find the appropriate value of a and b such that it is able to map every pair of (x,y) with highest accuracy and precision.

Once the algorithm predict the function using the training data-set, it will use that function to predict the future outcome.

Unsupervised Learning:

In this case, the program is provided with unlabeled data, and the algorithm explores the data and find some structure within. This is preferred when no historical data is available to provide as a training set. This works well on transactional data.

For e.g., it can identify groups of customers with similar attribute say based on their shopping patterns and then can be treated similarly in marketing campaigns or can be recommended items to buy based on the overall activity in that group.

Popular techniques include self-organizing maps, nearest-neighbor mapping, k-means clustering and singular value decomposition.

Reinforcement learning:

The machine train itself on a continual basis based on the environment it is exposed to and applies its enriched knowledge to solve business problems. This type of learning has three primary components: the agent (the learner or decision maker), the environment (everything the agent interacts with) and actions (what the agent can do).
A Reinforcement Learning(RL) essentially involves learning by interacting with an environment. An RL Agent learns from its past experience rather than its continual trial and error learning process as against supervised learning where an external supervisor provides example.