Machine learning and Artificial intelligence are the new buzz words that are being thrown around more than any other trending technology today. It is starting to reshape how we think about building products. It’s time we understood what it is and why it matters.
What is Machine Learning?
Machine Learning: (ML) is an area of computational science that enables machines (computers) to undertake tasks without being explicitly programmed. The idea behind machine learning is that by training computers to analyze and interpret existing data from prior human interactions, machines are able to find patterns and structures in data. This then enables machines to make decisions on data with similar characteristics without human intervention. Simply put, by exposing computers to an enormous amount of data, we allow machines to learn and then find recommendations and patterns when they see similar data in the future. Machine learning is already being used in numerous applications around us including chatbots, call centers, digital ads, airline pricing, etc.
To understand machine learning and why we need it, let’s take a step back and understand how business decisions were made before applying machine learning.
Rule-Based Problem Solving
A lot of traditional programs are written using a Rule-Based approach to problem-solving. For example, if you want to determine house prices in the San Francisco area programmatically, you would determine the key characteristics of the house that affect the price. These would be attributes like, lot size, square footage, neighborhood, school quality, etc. Then you would write a program to check each of these attributes and write if-then statements to determine the price of the house. A Rule-Based approach to solve problems works fine where there is a relatively small number of rules and finite sets of outputs.
Now imagine, if you introduce a new variable or the scenarios change and original rules do not apply. Now you have to go back and change the program. If there is an exception to the rule, you would have to update the program to account for the exception. One can easily see how this can go out of control for more complicated scenarios. This is where machine learning comes in.
Machine Learning vs Rule-Based Systems
Both machine learning systems as well as rule-based systems use data to make decisions. The key difference is that machine learning systems learn and adapt when they encounter new data with different characteristics without having to go back and change the system itself. So going back to the housing example, if over time, the demographics of the area change and new factors such as proximity to public transportation become important for buyers, the rules would have to be updated. However, in a machine learning system, the machine will learn this preference as new data is fed and can adapt to predict housing prices based on the new preference.
A few other things to clarify related to machine learning.
What is Artificial Intelligence
Artificial intelligence (AI) is a field that deals with teaching computers to mimic humans and human-like behavior. So its essentially adding capability to machines to think and understand like humans.
What is Deep Learning
Deep Learning is a particular method within machine learning that is really really good at one task. Deep learning algorithms use a method called Neural Networks and are trained to learn with a lot more data and make decisions regarding a particular task or problem.
Types of Machine Learning
There are three main ways in which machines can learn. The type of machine learning to deploy depends on the type of business problem one is trying to solve.
Supervised learning is a branch of machine learning where machines learn from past data that has known outputs for a set of inputs. This data where we have lots of input-output pairs are called labeled data. The idea here is that by analyzing lots and lots of labeled data, the model learns to generalize the output from labeled inputs so that when similar input data is fed to the model in the future, the machine is able to infer the output. The input fed to the machine can be one or more. In fact, the inputs can be finite or infinite. With today’s cloud computing and processing power, it is practically possible to feed 100s and 1000s of input parameters to the machine learning model and the machine is able to find patterns and develop hypotheses based on those inputs.
Supervised Machine Learning can be applied to two main types of problems.
Regression Problems are problems where output is continuous-valued. These algorithms predict the final value of the output. Some examples include predicting the home prices based on historical data, predicting how much to price the product based on sales data, predicting stock prices, or determine digital ad bid value. All of these have a numerical value for output given a set of inputs.
Classification Problems are a set of problems that have discrete outputs. These problems are centered around classifying an input in categories such as 0 or 1. But it does not have to be only two outputs. There can be multiple outputs such as classifying data as (0, 1, 2, 3). Some examples of classification problems are identifying an email as spam or not, predicting whether the Warriors are going to win or lose the next game, or even classifying a security threat as green, yellow, or red.
Unsupervised learning is a branch of machine learning where machines learn from unlabelled data and the model finds patterns and structure in the data without any help. This type of learning is usually applied to new data that has not been seen before and we have no information about how to interpret this data. The machine learning algorithms construct patterns or clusters from this data. There are two types of unsupervised learning.
Clustering is used to organize a large set of unlabelled data into clusters organized by one or more characteristics. These clusters can be used for social network analysis, market segmentation, weather data analysis, and machine learning models to try to identify similarities between data and group similar data together.
Association determines rules that govern large chunks of data and find linkages between the data. A good example is Amazon’s popular customers who buy A also buy B feature.
With reinforcement learning, the algorithm discovers through trial and error which actions yield the greatest rewards. 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). The objective is for the agent to choose actions that maximize the expected reward over a given amount of time. Common applications include robotics, gaming, and navigation.
Why Is Machine Learning Important?
With the advent of cloud technology, cheap storage, and connected devices, we have tons of data about user behavior and interaction with various systems. It is imperative that we use this data to make decisions about future products and improve the experiences we deliver to users. Data-driven decisions are increasingly becoming a part of the product development process already.
Machine learning takes it a step further by incorporating data into decision making and drive inferences, predictions, and outcomes based on data. Machine learning can be the key to unlocking the value of existing customer data and building products that learn from customer behavior and provide unique, personalized, and adaptive behavior that many customers today desire and expect.
Machine learning and Deep Learning technologies are currently being used in all types of industries including advertisements, healthcare, marketing, financial services, security, retail, and supply chain. There are numerous use cases in each of these verticals to incorporate machine learning into their products and decisions.
How are you using machine learning to solve a problem in your company? Please feel free to leave a comment.