Artificial intelligence and machine learning have emerged in the marketing industry as a pathway to competitive advantage. The best marketers are identifying, evaluating and testing AI-driven applications to make better sense of their data and assets, create personalized customer experiences and accelerate revenue growth. According to Forbes’ “10 Ways Machine Learning is Revolutionizing Marketing”:
- 84% of marketing organizations are implementing or expanding AI and machine learning in 2018.
- 75% of enterprises using AI and machine learning enhance customer satisfaction by more than 10%.
- 3 in 4 organizations implementing AI and machine learning increase sales of new products and services by more than 10% (Capgemini).
AI has the power to benefit multiple marketing initiatives, including brand engagement on social channels, content optimization for SEO, bot interactions and customer data analysis. With so many potential applications of Machine Learning, it’s important to understand the fundamentals of different types of machine learning algorithms and the mechanics behind their work.
In this series, we want to arm marketers with the basics they need to confidently tackle the machine learning landscape. To start, we’re taking you through the 3 main types of machine learning algorithms and explaining where you see them in daily life.
What is Machine Learning?
Machine Learning gives computers the ability to learn and continuously improve their performance of specific tasks without being explicitly programmed. It uses data and statistical techniques to “train” itself to make better predictions and decisions for any given task.
Essentially, machine learning allows computers to learn in the same way that humans do (with some restrictions, of course).
The 3 Types of Machine Learning
There are 3 primary types of machine learning algorithms:
- Supervised Learning
- Unsupervised Learning
- Reinforcement Learning
With Supervised Learning, we actively train the machine to provide specific outputs to a given input. We provide the machine with multiple examples of a given “input”, along with its correct “output” during the training phase, so that it can learn to recognize similarities and provide the right “output” on its own. Basically, we’re giving the machine the “right answers”.
For example, if we wanted a machine to recognize handwriting, we would show it a set of images of handwritten characters along with the correct labels for each of those characters. The computer learns the patterns and begins relating new images of handwriting to the correct characters on its own.
There are two subgroups in Supervised Learning:
Regression: This is used when the output will be a continuous quantity, such as salary, age or height. For example, how tall a child will be as an adult.
Classification: This is used when the output will be a category, such as “spam” or “not spam” in your email inbox. This is commonly used for identity fraud detection, diagnostics and even image classification.
In Unsupervised Learning, we don’t provide the machine with any output categories or labels. The machine uses the data we input to find rules, detect patterns and summarize them so that we can develop meaningful insights. Instead of receiving explicit feedback on if a given output is correct or not, the machine learns by analyzing the presence or absence of similarities in new inputs.
For example, targeted marketing applications and recommender systems use unsupervised learning to target users with relevant content. We see this when Netflix recommends a TV series. It identifies users with similar preferences or viewing habits and serves content that other like-minded users have enjoyed.
Clustering is one of the more prominent types of unsupervised learning, which involves grouping similar “inputs” together and analyzing new data based on these “clusters”.
Reinforcement Learning is newer than supervised and unsupervised learning. We don’t provide the machine with the data and the “right answers”, but we provide a method for the machine to quantify its performance in the form of a reward signal. In this form of learning, the machine decides on an output, given its current state, and they will either receive negative feedback or a reward signal. Over time, the machine will be able to determine the optimal behavior to maximize its chances for a reward signal.
For example, there’s a reinforcement learning model that was trained to play Atari video games using only the pixel output from the game as input. The model was eventually able to outperform human players.
This form of learning closely resembles how humans and animals work – we learn by trying different things and then getting rewarded when we do something well.
Machine Learning in Daily Life
Artificial intelligence and machine learning can seem overwhelming, but it’s already being used in most people’s everyday life.
Here are some examples of how machine learning is impacting your day-to-day:
- Google Assistant, Apple Siri and Alexa
- Google Maps Traffic Predictions
- Facebook Friend Recognition
- Gmail’s Priority Inbox
- Spotify’s Discover Weekly
- Amazon Recommended Products
To understand more about machine learning and, more specifically, it’s impact on digital asset management, be sure to watch Henry Stewart Events’ webinar, “A practical approach to artificial intelligence using cognitive metadata in DAM.”