Welcome to this executive briefing on Reinforcement Learning.
I appreciate that your time is valuable. And that’s why in this course, we will only cover the ultimate essentials for you to get up to speed with reinforcement learning and see tangible examples of how this technology can add substantial value to your business.
Here’s how this course is structured:
- First, we will talk about what reinforcement learning is and why it has such a high potential to disrupt virtually any business and any industry
- What we can take away from that. Then we will discuss the six core advantages of reinforcement learning that make it one of the most cutting-edge technologies of today
- Benefited the most from these new technologies in the recent years
- Finally, we’ll look at 10 different case studies of reinforcement learning already being applied in different companies and industries
Here we are after the real results, the numbers, so that we can get a feel for what reinforcement learning can do for our businesses in a tangible way.
So, I hope you’re as excited as I am about upcoming tutorials.
And without further ado, let’s dive into the world of reinforcement learning.
Table of Contents
- What is Reinforcement Learning?
- Reinforcement Learning in Humans
- Advantages of Reinforcement Learning?
- Reinforcement Learning in Marketing – 5 Applications
- 10 Use-cases – with Real Results
- Summary
What is Reinforcement Learning?
Reinforcement learning is an area of machine learning where an imaginary agent or the computer in our case is being presented with a problem and is being rewarded with a plus one (+1) for finding a solution to the problem or punished with a minus one (-1) for not finding a solution.
Well, there are three main groups of algorithms in machine learning, unsupervised learning, supervised learning, and reinforcement learning, it can be very valuable to appreciate the differences between the three.
- Unsupervised learning is used for discovery of new patterns, for example, clustering of customers into groups based on their similarities
- The core principle here is that the resulting groups did not exist prior, but rather are suggested by the machine in the process
- Supervised learning, on the other hand, is when we teach a machine to search and identify patterns that we have seen before
For example, classification of pictures of dogs and cats into two categories. Dogs and Cats is an example of a supervised learning algorithm in action.
First, we show the algorithm thousands of already labelled images so it can extract the features that are essential to dogs and the features that are essential to cats.
After this, the algorithm will be able to categorize new images as either dogs or as images of cats. The difference of this approach is to unsupervised learning is that we must first provide labelled data for the algorithm to learn.
Finally, reinforcement learning is something entirely different.
Unlike with supervised learning, the agent is not given instructions on how to perform the task. Instead, it performs a random action and interacts with its environment.
So that, in a nutshell, is what reinforcement learning is all about.
And I hope you’re super excited to explore this field further, because this is indeed the future of machine learning.
Reinforcement Learning in Humans
Reinforcement learning is surprisingly like real life, we humans use it all the time.
Let’s look at an example of a baby learning to walk when it’s starting to learn to walk, and it falls over, it hurts itself and its nervous system sends a pain signal into the brain.
- Pain is a concept that is generated inside our body. It doesn’t exist outside of our minds. It’s nothing more than an electrical signal that’s going into your brain.
- And so, what happens is that the baby falls over and understands that I’m experiencing pain or subconsciously it feels pain and therefore it learns not to do that action again
- For instance, it was standing on if it’s on its left leg and it’s raised its right foot towards its shoulder or something like that and made it fall over
- And that pain signal is a reward, a negative reward for the baby’s mind, for this baby’s subconscious not to ever do or try to avoid that action because it leads to pain
- On the other hand, if the baby was able to make a few steps forward and grab a shiny object and it experienced joy, then joy in that case is a positive reward
- It’s a reward that is giving the baby positive reinforcement to continue doing more of those actions. And therefore, the baby will learn that by doing those actions, I get positive rewards, so I’ll do more of them
That’s how a baby learns to walk.
Reinforcement learning is based on the same principles. And because of that, because it is so like the way our own intelligence works, out of all the existing algorithms, reinforcement learning is the closest thing that we have gotten to in terms of true artificial intelligence.
That’s what we’re going to be talking about in the upcoming tutorials.
Advantages of Reinforcement Learning?
Here’s a quote from John Langford:
There are many situations where you just can’t label data effectively. You must learn from rewards. And since supervised learning relies on large, labelled data sets, reinforcement learning has a much higher scope of application than any form of supervised learning.
That’s reinforcement!
Reward System
Learning doesn’t require large, labelled data sets. It’s a massive advantage because as the amount of data in the world grows, it becomes more and more costly to label it for all required applications. Here are some other advantages of reinforcement learning advantage.
It’s Innovative
Unlike reinforcement learning, supervised learning is imitating whoever provided the data for that algorithm. The algorithm can learn to do the task as well or better than the teacher but can never learn a completely new approach to solving the problem.
On the other hand, reinforcement learning algorithms can come up with entirely new solutions that were never even considered by human’s advantage.
Bias Resistance
If there’s bias in the way the data is labelled, then a supervised learning algorithm will pick up that bias and learn inherited bias. In this sense, reinforcement learning algorithms are better tools to find solutions free from bias or discrimination advantage than before is online learning.
Reinforcement Learning learns in real-time
Reinforcement learning combines exploration when the machine tests new approaches on the flight to find better solutions and exploitation.
When the machine exploits the best solutions which it has found thus far, this means that it can bring results while improving. At the same time, other algorithms would require retraining and redeployment. To accomplish that, reinforcement learning just keeps going.
It’s Goal Oriented
Reinforcement learning can be used for sequences of actions.
While supervised learning is mostly used in an input output manner, reinforcement learning can be used for tasks with objectives such as robots playing soccer or self-driving cars getting to their destinations, or an algorithm maximizing return on investment, on ad spending
Reinforcement learning is adaptable
Unlike supervised learning algorithms, reinforcement learning doesn’t require retraining because it adapts to new environments automatically on the fly.
Those were 6 advantages of reinforcement learning.
Knowing these advantages can help you decide if reinforcement learning is the right tool for the artificial intelligence applications that you have in mind for your organization in your industry.
Reinforcement Learning in Marketing – 5 Applications
Because of its online learning advantage or the ability to auto correct on the fly, reinforcement learning is most popular.
Current business application is perhaps, MARKETING!
Virtually any business could benefit from cutting edge technology in its marketing. And that’s why in this article, we will look at 5 examples of how reinforcement learning will revolutionize the way we do marketing, even if you don’t apply these methods right away. They are a good set of tools to keep in the back of your mind for the future
Creating personalized recommendations
Reinforcement learning can dynamically consider customer specific preferences, needs and behaviours to ensure they get high quality recommendations that resonate with them.
This is effectively like a recommender engine, which you would find at Netflix, but with the added benefit that it can learn real time and adjust its recommendations real time based on the success that is having. It is very relevant in the case of ecommerce, where thousands of new items are appearing every hour and hundreds of thousands of customers are shopping every day.
It is very relevant in the case of ecommerce, where thousands of new items are appearing every hour and hundreds of thousands of customers are shopping every day
Optimizing advertising budgets
Understanding which advertisements bring the best return on investment is incredibly complex, often impossible to achieve with conventional means.
Not only are you trying to cross-reference effectiveness data with various factors that influence each individual customer, you’re also trying to do that in real time all the time.
The algorithm matches known user preferences and the context of the advertisement to rank each advertising slot and determine the optimal did.
Selecting the best content for advertisements
Reinforcement learning blows, AB testing out of the water. AB testing is static.
You must wait until the end of the test to see the results, whereas reinforcement learning works on the fly, which means that it can find out the most optimal content much quicker and start showing it to the customer sooner.
This minimizes the number of times non optimal content is displayed, thereby maximizing revenue. That’s one of the reasons why Baidu, the Chinese equivalent of Google, deployed a deep reinforcement learning algorithm called Moonrise, which led to significant improvements in search, relevance, and ad performance.
Increasing customer lifetime value
Reinforcement learning allows a company to focus on optimizing the lifetime value of a customer rather than short term revenue results.
Adobe researchers proposed an algorithm to display personalized ads to existing customers. The algorithm can analyze the customer’s perceived preferences to ensure they aren’t overwhelmed by advertisements for services they don’t need.
While that might have decreased the number of clicks in the short term, the comprehensive approach nurtures a customer and maximizes their lifetime value use
Predicting customers responses to price plan changes
By using inverse reinforcement learning, we can observe consumer behavior and estimate what their reward function looks like.
In other words:
- What goals are they pursuing when shopping?
- Are they trying to save money or buy the high-end brands?
- What quality to price ratio is optimal for them?
- What level of service do they find optimal?
- What features are most desirable?
Once this information is obtained, the company can use it to innovate its products or marketing, so they will go.
Those are 5 examples of how reinforcement learning is disrupting marketing already and is going to disrupt marketing into the future.
And hopefully that gave you some ideas of how you can apply reinforcement learning in the marketing parts of your business.
10 Use-cases – with Real Results
In this section, we’re going to look at 10 examples of reinforcement learning used in action by companies today to achieve real results, real tangible results
Google and their creation AlphaGo
In the famous example, AlphaGo learned to play the game of Go, which is considered to be more complex by orders of magnitude than the game of chess.
For example, by playing games against itself and using reinforcement learning with no outside assistance whatsoever, AlphaGo was able to truly master the game. It played 60 games against the top players of the world and 160 to zero.
This artificial intelligence has become virtually unbeatable by anything or anyone other than itself, and that is a pinnacle of reinforcement learning at the moment.
Google, In Space of Energy Management
Google implemented DeepMind, a system of neural networks trained on different operating scenarios and parameters to reduce the amount of energy they use for cooling their service centers by up to 40% , as you can imagine, that’s a massive energy saving for Google.
Decision service
Uses reinforcement learning in advertising.
So, Decision Service is a contextual learning service to improve existing advertising systems. It uses reinforcement learning to achieve a click through rate improvement of 25% – 33% and a revenue lift of 8% just by adding reinforcement learning example.
Trendyol
Uses reinforcement learning in email advertising.
So, Trendyol is an email automation tool that distinguishes which messages will be most relevant to which customers with the help of reinforcement learning they were able to achieve 30% lift in clickthrough rates, 60% lift in response rate, and 130% lift in conversion rates.
Alibaba
Uses reinforcement learning for advertising display bidding.
Alibaba developed a system to quantify how likely a customer is to click on a particular ad based on their preference, the context, and the ad itself.
This ability to calculate how much each advertisement is worth allowed them to increase the return on investment by 240% percent without increasing the advertising budget.
Electa
Uses reinforcement learning for Energy Management.
So, scientists from Elector, the largest research group on electrical energy systems in Benelux, developed an a reinforcement learning system to optimize hot water control systems, applying this invention to a set of 32 houses.
It reduced energy consumption by around 22% with no loss of comfort reported by the occupants, and that’s a massive progress, not just in terms of cost savings, but also it has an environmental impact.
If we use 20% less energy, that’s going to benefit the world.
Fanuc / Tesla
Uses Reinforcement Learning for manufacturing and robotics.
So, robotics giant Fanuc develops robots that quickly learn to perform new tasks, including soring products or delivering them to the right place or person. That might sound a bit boring, but Tesla actually uses one hundred and sixty of these robots in their factory.
And what the robots can accomplish is using reinforcement learning. These robots can achieve 90% accuracy on new tasks overnight. So is almost the same as if an expert were to program them. But the difference is that you don’t have to program at all.
Just tell them what goal they need to achieve. And overnight they will learn how to accomplish themselves independently.
Unnamed Company
Uses Reinforcement Learning for inventory management.
So, inventory management is a delicate process of seeking the balance between keeping enough stock in the warehouse to ensure the business keeps moving and making sure that the stock doesn’t drain the business of its cash reserves.
They incorporated a reinforcement learning powered inventory management system saw a 32% reduction in costs across operation and not just speaking of cost reduction.
Consider the also environmental ethical implications that how much extra food to supermarkets, for example, stock them as opposed to how much they sell using reinforcement learning those stocks can be optimized
Cambridge University
Uses reinforcement learning in healthcare.
Due to the nature of the field, artificial intelligence has mostly entered healthcare in the form of supervised learning, intending to replicate the work of medical professionals as closely as possible. However, recent work with reinforcement learning has shown that it’s possible to improve on the existing method.
Researchers from Cambridge developed a reinforcement learning algorithm that improves treatment policies for patients with sepsis.
Right now, we might find it difficult to follow the recommendation of a computer program over that of a doctor. These algorithms are already proven to be more accurate than doctors and will soon be saving lives.
So ultimately, the future of medicine is also in the hands of reinforcement learning because, as we discussed before, it can be more innovative and come up with methods that humans have haven’t even thought of.
Tesla, Google, and Other Companies
Uses reinforcement learning for self-driving cars.
So, one of the most hyped-up areas where reinforcement learning is bound to make a big splash is enabling self-driving cars with Tesla, Google and other major players entering the arena.
It is predicted that by 2040, 95% of all new vehicles sold will be fully autonomous. Not only is this an enormous market, but it will also help dramatically improve our quality of life. We currently spend about three years of our lives in transit and one point 25 million lives are lost every year due to traffic accidents.
Accidents which can largely be avoided by a powerful algorithm that doesn’t get tired, doesn’t make mistakes, sees everything, and reacts within milliseconds.
So, there we go with top 10 reinforcement learning use cases.
Those were ten examples of reinforcement learning and action bringing tangible results to businesses already today.
I hope this inspires some ideas of how you can use reinforcement learning in your industry and in your business.
Summary
So in a nutshell,
Reinforcement Learning (RL) is a branch of artificial intelligence that focuses on training agents to make decisions in an environment by learning from the consequences of their actions.
Future trends in RL include increased use of deep reinforcement learning, which combines deep learning techniques with RL algorithms to allow agents to learn from high-dimensional observations.
There is also a growing interest in using RL in real-world applications such as autonomous driving, energy management, and healthcare. As the field continues to evolve and advances are made, it is possible that RL will become increasingly integrated into various industries and have a more significant impact on society. However, like any technology, it is important to consider the ethical and safety implications of its use and ensure that it is used responsibly.
So, there we go.
I really hope that now you are much more comfortable and confident speaking about reinforcement learning and more importantly, seeing how you can apply them in your industry and in your business.
Thank you for reading our blog, we hope you found the information provided helpful and informative. We invite you to follow and share this blog with your colleagues and friends if you found it useful.
Share your thoughts and ideas in the comments below. To get in touch with us, please send an email to dataspaceconsulting@gmail.com or contactus@dataspacein.com.
You can also visit our website – DataspaceAI
One thought on “What is Reinforcement Learning? Top 10 most effective business use cases”