Covering the basics of Machine Learning

My goal with Tech Tuesday and this article is to simply provide more information about emerging technologies to every day people, every Tuesday, using simple terms and breaking down the technology into 3 categories: 1) What is it? 2) How does it work? and 3) What are its current and future applications?

So this week, I'm going to cover the topic of Machine Learning!

Lets jump right into it:

What is it?

We, as humans have the unique ability to reflect and learn from our past experiences and act based on that knowledge, without having to be told explicitly to do so. Computers on the other hand, are strict logic machines and unfortunately do not work this way. This means that if we want them to do something, they have to be provided with step-by-step instructions on exactly what to do through programmed code and scripts, which you could imagine would end up being somewhat inefficient. This is where Machine Learning comes into play:

Machine learning is an application of artificial intelligence (AI) that provides systems the ability to automatically learn and improve from experience without being explicitly programmed to do so, with the primary aim being to allow computers to learn automatically, without human intervention and perform actions accordingly. It focuses on the development of computer programs that can access data and use it to learn for themselves, through provided observations, examples, direct experience, or instruction, in order to look for patterns in previous data and execute, based on the examples that we provide. 

How does it work?

The working process of Machine Learning is divided into 3 major categories: Supervised Learning, Unsupervised Learning and Reinforcement Learning

1) Supervised Learning

Supervised learning refers to a type of machine learning where the computer is provided with a "training set" of data, which includes input and output variables, and the goal of the computer is to determine the inferred relationship or "function" between the values through a learning algorithm. The functions are of the form: y = F(x), where "x" is the input variable, "y" is the output variable and "F" is the function or relationship between the two variables. An example of such a function would be: y = 2x, where the output values are always 2 times the input value. The task of the computer would be to determine this function by analyzing the input and output values, through a learning algorithm. This process is called Supervised learning because the process of an algorithm learning from the training data-set can be thought of as a teacher supervising the learning process. Here, we know the correct answers, but the algorithm makes repeated predictions from the training data and is corrected by the teacher, until it can determine a suitable relationship for the values. This process can then be applied to a situation where the computer can use what has been learned from the past to new data, using labeled examples, to predict future outcomes.

Supervised learning problems can be further categorized into: regression and categorization problems

• Regression problems: Refers to the scenario where the output variable is a real value, such as "height" or "dollars", etc.
• Categorization problems: Refers to the scenario where the output variable is a category, such as "black", "blue", "disease", etc.

2) Unsupervised Learning

Unsupervised learning refers to a type of machine learning where the computer is just provided with input variables "x", and no output values. The goal for the system is to analyze and model the underlying structure or distribution of the data being provided in order to learn more about it. Unlike supervised learning, there are no correct answers or guidance here, the algorithms are just expected to analyze and discover any potential structure or model within the data, to get more insight into it. Unsupervised learning problems can be further categorized into: clustering and association problems

• Clustering: A clustering learning problem refers to a scenario where you would want to discover any inherent groupings in the data provided. An example would be grouping customers by their purchasing behavior, or grouping them by a food preference (vegan, non-veg), etc.
• Association: An association-rule learning problem refers to a situation where you would want to discover the association rules that describe the majority of your data. An example would be learning that people that buy "A", would also probably buy "B".

3) Reinforcement Learning

The Reinforcement learning model focuses on learning through a trial and error, delayed reward based search method, with the goal to maximize and optimize the performance of a given problem or system. Given a goal, the algorithm performs actions and examines the result of those actions to determine the most effective approach to achieve the desired goal. If the result of an action is "dull" or ineffective, the computer deems it as a "negative reward" or error, and takes steps to not perform such an action again, and conversely if the result of an action is effective, the computer focuses on maximizing similar actions in the long term, to get the most ideal performance. A simple way to think of this is how you would reward your child for doing something good, reinforcing that behavior and give them a form of punishment for doing something bad, so that they learn to not do those actions again. The algorithms perform in a "delayed reward" environment, which means it can be difficult to determine which action leads to which result, so it is a process that slowly performs better and better with more time, exposure and rewards processed with a given problem.

A helpful diagram of the 3 types of machine learning summarized, is shown below:

What are its current/future applications?

Machine learning is being used in a wide range of industries today, due to it's strong processing and optimization capabilities. As seen in the picture above, the different categories of machine learning have different applications to the real world, each with its own benefits.

Supervised learning can be used in a variety of industries, with the forecasting and prediction capabilities used in advertising campaigns, market forecasts, weather predictions, as well as biological applications through bio-informatics, which is the storage of our biological information like fingerprints and iris details which are widely used in smartphones for authentication and wellness purposes, along with speech recognition for similar uses as well. The classification capabilities are applicable for things such as spam and fraud detection through emails, Customer Relationship Management (CRM) applications to analyze emails and communications, network and systems diagnostics to troubleshoot through difficult problems.

For unsupervised learning, clustering algorithms can be applied in consumer-analysis based settings where a company, such as a retail clothing store would want to gain insight into and analyze purchasing behaviors of customers, targeted marketing by major brands to consumers through social media for products/services, with a common example being through the Facebook News Feed. Association algorithms can be used in conjunction with the above, with applications to the retail and consumer space for purchasing behaviors and market segmentation based on different consumer demographics such as age, gender, income, etc.

Reinforcement learning is applied currently in avenues such as robotics; analyzing video images and linking to them to the robot actions, traffic light control systems to reduce congestion on roads, business management for e-commerce recommendations; which companies are focusing on increasingly, gaming AI optimization; in games such as Poker and Dota 2, the finance sector for risk management, pricing and trading, STEM fields for research applications, etc.

Machine learning is incredibly powerful to gain insight, analyze data and determine the most efficient and effective solutions to our problems. There is an increasing interest into AI and machine learning algorithms for companies, with an estimated $5 billion dollars invested in 2017. In the future, we can expect to see more demand and understanding of the applications of machine learning to optimize all forms of data processing and ultimately improve the impact of our technologies and ideas.

.

.

.

Thank you for taking the time to read this article, I hope it was a helpful and insightful read into the world of Machine Learning! I by no means am an expert in technology, just simply someone with a passion for it and a love for sharing information with others!

If you have any tips/ideas for the Tech Tuesday format, or other topics to cover for next week's edition, please let me know!

Cheers.