One of the first things to know about machine learning is that you will be working with one of three types of algorithms: supervised learning, unsupervised learning and reinforcement learning. Here’s a quick summary of the three types:
Let’s first take a quick step back and look at situations in which supervised or unsupervised learning would be useful in developing an AI strategy: You would probably use some form of supervised learning (e.g., linear regression, random forest, support vector machines) to solve problems such as customer segmentation, churn prediction, likelihood to purchase or fraud detection.
And you would probably use some form of unsupervised learning (e.g., clustering, k-means, neural networks) when you’re working on problems like facial recognition, language translation or speech analysis.
In those kinds of situations, you already have a pretty good idea of the data you have, what’s going on and how to solve the problem. You’re using machine learning to find interesting patterns in that data to get to a better solution, accelerate the process and get to your solution faster. You have a ton of data and you want a machine to find interesting patterns -- and tell you what choices to make based on what it discovers.
Another popular type of AI, reinforcement learning is a form of supervised learning, but only given partial information. RL is an increasingly popular technique for organizations that deal regularly with large complex problem spaces. Because RL models learn by a continuous process of receiving rewards and punishments on every action taken, it is able to train systems to respond to unforeseen environments. Industrial systems, including supply chain management and industrial robotics, are good examples of large problems well spaces perfectly suited to be solved with reinforcement learning.
With reinforcement learning, domain experts and organizations typically know what they want a system to do -- but they want to automate or optimize a specific process.
In the case of supply chain management, it would be extremely difficult, if not impossible, to write a program that could effectively manage every possible combination of circumstances occurring in everyday scenarios. The trucks could break down, the food could spoil, bad weather could force road closures – the list of potential hazards is virtually infinite.
Here’s something else to consider in the supply chain example: The problem space is highly dynamic; it’s constantly changing. Every decision made by your system has an impact on the world and team around it.
As a result, your system must be highly adaptive. Again, this is where reinforcement learning techniques are especially useful since they don’t require lots of pre-existing knowledge or data to provide useful solutions.
Applied Reinforcement Learning
Deep reinforcement learning is at the cutting-edge right now, with many of the world’s best researchers working on improving the core algorithms. But it’s finally reached a point that it can be applied to real-world industrial systems. Organizations should implement reinforcement learning in their AI strategies when:
To see reinforcement learning in action, watch a demo of the Bonsai Platform using RL to teach a robotic arm how to intelligently grasp and stack blocks in simulation.
You can also download our free whitepaper, “AI for Industrial Applications” to learn more about how reinforcement learning can solve real-world enterprise problems.