Welcome to Demystifying Artificial Intelligence, a blog post series where we will look at Artificial Intelligence through the lens of practical business applications. We introduced this blog series in the previous post. Today, we will discuss Machine Learning.
What is Machine Learning?
Machine Learning technology allows machines to learn and improve from experience (data) without being directly programmed. Developers feed the machine huge amounts of data, and it can learn patterns and relationships that it applies to understand new pieces of data streaming into the system.
It’s important to remember that Machine Learning and Artificial Intelligence are not synonyms. Machine Learning is only one of the important specializations that contribute to the broader field of Artificial Intelligence.
Depending on the problem you need to solve, different types of Machine Learning are used. Let’s discuss 3 popular types of machine learning.
3 Types of Machine Learning
1. Deep Learning
Deep Learning is a type of machine learning that uses neural networks to solve problems. These networks have different “layers” that learn to identify features, which are used to classify the data. This technology is effective for tasks such as image and speech recognition.
Application: Cat Picture Detection
So, as an example, a deep learning network may be trained to identify pictures of a cat. One layer might learn to identify pointy ears, and another layer might learn to identify the appearance of fur. Now some dogs may have pointy ears, and some cats may be furless, but with enough data, the network can learn when it has identified the critical features that make a photo most likely to contain a cat.
Application: Fraud Detection
In the same way, a network could be trained to identify fraud for a financial institution. Certain transactions can indicate to a system that fraud is taking place the same way as pointy ears indicate a cat. The deep learning algorithm learns from data such as the company’s past transactions and fraud cases, customer demographics, and behavior. It would then be able to detect if something is unusual, and flag it for human analysis. In result, the institution can reduce losses to the firm and customer and improve security.
2. Reinforcement Learning
Reinforcement Learning is a type of machine learning that uses agents that use a “trial and error” to develop a policy. The agent interprets the data and make decisions that have a negative or positive score. Over many different trials they learn what decisions led to an optimal result. It is based on the same theory as Pavlov’s Dog, the classic experiment from Behavioral Psychology where a dog learns to associate food with the sound of a bell.
Application: Price Optimization
For example, you can build an agent that optimizes prices for your product. It will learn from data from the company’s sales and customer behavior, market trends and competitor prices. The agent receives positive or negative feedback based on how much its pricing decisions impact the company’s profit. Over time, it learns to set optimal decisions which makes the business more profitable and competitive.
Application: Drug Discovery
Another application is drug discovery for a pharmaceutical company. Such an agent learns to identify patterns and regularities in the outcomes of clinical trials. It flags certain compounds or molecules as potential new drugs and therapies. The agent helps scientists more quickly and effectively determine promising leads, possible side effects, and potential mechanisms of disease.
3. Incremental Learning
Incremental Learning is a type of machine learning that can improve itself over time. Most types must be completely retrained from scratch when new data is added, but Incremental Learning can continuously learn from data. This type of learning is especially important when data may change, such as robotics and online systems.
Application: Personal Assistant
Consider a system that has the purpose of being a personal assistant. As the user interacts with the system, it learns more about that user’s preferences and can optimize scheduling meetings or activities at the perfect time for that user. Such an assistant can reduce stress for the user, who is freed to focus on their life and work.
Application: Chat Bot
Incremental learning systems are also useful in creating chatbots. Chatbots learns to identify common customer requests from existing data on customer’s interactions with the company. Then, the more it interacts with customers, the better it becomes at replying.
Another application is collections efforts. These systems initially learn from past collection efforts and customer demographics. The system tries to prioritize its strategies based on what a person is most likely to respond to, such as payment reminders, phone, letters, or email. Over time, it continuously improves its collection efforts, so that a company is better able to collect outstanding debts.
We hope this helped you understand what types of machine learning are best for certain problems. Come back Thursday, December 15th for the next one. We will discuss the role of Data Science in Artificial Intelligence.
If you want to be notified, please subscribe for updates!
Satish Medapati – Head of AI and Data Solutions
Melanie Allen – Product Marketing Content Writer
Are you looking to hire Machine Learning Engineers? This Hiring Guide from Toptal offers best practices, advanced techniques, and interview questions and answers. It lays out the crucial skills and knowledge that a premium ML Engineer should have.