8 min read

Mckinsey report predicts that artificial intelligence techniques including deep learning and reinforcement learning have the potential to create between $3.5 trillion and $5.8 trillion in value annually across nine business functions in 19 industries.

Reinforcement learning (RL) is an increasingly popular technique for enterprises that deal with large complex problem spaces. It enables the agents to learn from their own actions and experiences. When working in an interactive environment, they use a trial and error process to find the best-optimized result. Reinforcement learning is at the cutting-edge right now and it’s finally reached a point that it can be applied to real-world industrial systems.

We recently interviewed Sudharsan Ravichandiran, a data scientist at param. ai, and the author of the book, Hands-On Reinforcement Learning with Python. Sudharsan takes us on an insightful journey explaining to us why reinforcement learning is trending and becoming so popular lately. He talks about the positive contributions of RL in various research fields such as gaming industry, robotics, inventory management, manufacturing, and finance.

Author’s Bio

Sudharsan Ravichandiran, author of the book, Hands-On Reinforcement Learning with Python is a data scientist, researcher, and YouTuber. His area of research focuses on practical implementations of deep learning and reinforcement learning, which includes natural language processing and computer vision.

He used to be a freelance web developer and designer and has designed award-winning websites. He is an open source contributor and loves answering questions on Stack Overflow. You can follow his open source contributions on GitHub.

Key Takeaways

  • Reinforcement learning adoption among the community has increased exponentially because of the augmentation of reinforcement learning with state of the art deep learning algorithms.
  • It is extensively used in the Gaming industry, robotics, Inventory management, and Finance. You can see more and more research papers and applications leading to full-fledged self-learning agents.
  • One of the common challenges faced in RL is safe exploration. To avoid this problem, one can use imitation learning (learning from human demonstration) to provide the best-optimized solution.
  • Deep meta reinforcement learning will be the future of artificial intelligence where we will implement artificial general intelligence (AGI) to build a single model to master a wide variety of tasks. Thus each model will be capable to perform a wide range of complex tasks.
  • Sudharsan suggests the readers should learn to code the algorithms from scratch instead of using libraries. It will help them understand and implement complex concepts in their research work or projects far better.

Full Interview

Reinforcement learning is at the cutting-edge right now, with many of the world’s best researchers working on improving the core algorithms. What do you think is the reason behind RL success and why RL is getting so popular lately?

Reinforcement learning has been around for many years, the reason it is so popular right now is because it is possible to augment reinforcement learning with state of the art deep learning algorithms. With deep reinforcement learning, researchers have obtained better results.

Specifically, reinforcement learning started to grow on a massive scale after the reinforcement learning agent, AlphaGo, won over the world champion in a board game called AlphaGo. Also, Deep reinforcement learning algorithms help us in taking a closer step towards artificial general intelligence which is the true AI.

Reinforcement learning is a pretty complex topic to wrap your head around, what got you into RL field? What keeps you motivated to keep on working on these complex research problems?

I used to be a freelance web developer during my university days. I had a paper called Artificial Intelligence on my Spring semester, it really got me intrigued and made me want to explore more about the field. Later on, I got invited to Microsoft data science conference where I met many experts and learned more about the field way better. All these got me intriguing and made me to venture into AI.

The one thing which motivates and keeps me excited are the advancements happening in the field of reinforcement learning lately. DeepMind and OpenAI are doing a great job and massively contributing to the RL community. Recent advancements like human-like robot hand control to manipulate physical objects with unprecedented dexterity, imagination augmented agents which can imagine and makes decisions, world models where the agents have the ability to dream excite me and keeps me going.

Can you please list down 3 popular problem areas where RL is majorly used? Also, what are the 3 most current challenges faced while implementing RL in real-life? As a developer/researcher how you are gearing up to solve them?

RL is predominantly used in the Gaming industry, robotics, and Inventory management. There are several challenges in Reinforcement learning. For instance, safe exploration.

Reinforcement learning is basically a trial and error process where agents try several actions to find the best and optimal action. Consider an agent learning to navigate/learning to drive a car. Agents don’t know which action is better unless they try them. The agent also has to be careful in not selecting actions which are harmful to others or itself, say, for example, colliding with other vehicles.

To avoid this problem, we can use imitation learning or learning from a human demonstration where the agents learn directly from the human supervisor. Apart from these, there are various evolutionary strategies used to solve the challenges faced in RL.

There are few positive developments in RL happened from Open AI and DeepMind team that have got widely adopted both in research and in real-world applications. What are some cutting-edge techniques you foresee getting public attention in RL in 2018 and in the near future?  

Great things are happening around RL research each and every day. Deep Meta reinforcement learning will be the future of AI where we will be so close to achieving artificial general intelligence (AGI). Instead of creating different models to perform different tasks, with AGI, a single model can master a wide variety of tasks and mimics the human intelligence.

Gaming and robotics or simulations are the two popular domains where reinforcement learning is extensively used. In what other domains does RL find important use cases and how?


In manufacturing, intelligent robots are used to place objects in the right position. If it fails or succeeds in placing the object in the right position, it remembers the action and trains itself to do this with greater accuracy. The use of intelligent agents will reduce labor costs and result in better performance.

Inventory management

RL is extensively used in inventory management, which is a crucial business activity. Some of these activities include supply chain management, demand forecasting, and handling several warehouse operations (such as placing products in warehouses for managing space efficiently).

Infrastructure management

RL is also used in infrastructure management. For an instance, Google researchers in DeepMind have developed RL algorithms for efficiently reducing the energy consumption in their own data center.


RL is widely used in financial portfolio management, which is the process of constant redistribution of a fund into different financial products and also in predicting and trading in commercial transactions markets. JP Morgan has successfully used RL to provide better trade execution results for large orders.

Your recently published ‘Hands-On Reinforcement Learning with Python‘ has received a very positive response from the readers. What are some key challenges in learning reinforcement learning and how does your book help them?

One of the key challenges in learning reinforcement learning is the lack of intuitive examples and poor understanding of RL fundamentals with required math. The book addresses all the challenges by explaining all the reinforcement learning concepts from scratch and gradually takes readers to advanced concepts by exploring them one at a time. The book also explains all the required math step by step intuitively along with plenty of examples.

My intention behind adding multiple examples and code to each chapter was to help the readers understand the concepts better. This will also help them in understanding when to apply a particular algorithm.

This book also works as a perfect reference for beginners who are new to reinforcement learning. Are there any prerequisites needed to get the most out of the book? What do you think they should keep in mind while developing their own self-learning agents?

Readers who are familiar with machine learning and Python basics can easily follow the book. The book starts with explaining reinforcement learning fundamentals and reinforcement learning algorithms with applications and then it takes the reader in understanding deep learning algorithms followed by the book explaining advanced deep reinforcement learning algorithms. While creating self-learning agents, one should be careful in designing reward and goal functions.

What in your opinion are the 3-5 major takeaways from your book?

  • The book serves as a solid go-to place for someone who wants to venture into deep reinforcement learning.
  • The book is completely beginner friendly and takes the readers to the advanced concepts gradually.
  • At the end of the book, the readers can master reinforcement learning, deep learning and deep reinforcement learning along with their applications in TensorFlow and all the required math.

Would you like to add anything more to our readers?

I would suggest the readers code the algorithms from scratch instead of using libraries, it will help them in understanding the concepts far better. I also would like to thank each and every reader for making this book a huge success. My best wishes to them for their reinforcement learning projects.

If you found this interview to be interesting, make sure you check out other insightful articles on reinforcement learning:

Subscribe to the weekly Packt Hub newsletter. We'll send you the results of our AI Now Survey, featuring data and insights from across the tech landscape.

* indicates required