Reinforcement learning is more and more relevant for the industry!

published: 2021-12-12
3 minutes read

Virginie Marelli

Reinforcement learning is more and more relevant for the industry!

The AI environment is becoming increasingly mature. Once hard to crack processes are now being automated in no time and it won’t be long before we even start automating these automations on a large scale. However, the complexity of AI challenges is also evolving. AI experts now sink their teeth into fully automating complex processes such as production lines and concepts like dynamic pricing. They are therefore increasingly turning to optimisation methods such as reinforcement learning and applications that, in turn, open doors to more complex, better and faster possibilities.

Reinforcement learning (RL) offers a particularly powerful way of optimising processes. It is a concept that allows different objectives to be bundled into one model. Just like in a video game, you play a level not only to get to the endzone in one piece, but also to collect as many coins as possible. Reinforcement learning is a powerful leap forward in AI because it allows entities or quantities in processes to be optimised with a clearly stated goal in mind, but without having to have a defined path in mind to reach that goal.

Learning through trial and error

In reinforcement learning, machine learning models are trained to make sequential decisions. In its environment, the model uses trial and error to reach a goal and arrive at a solution. Each action the model takes is rewarded or punished by the programmer, who thereby pushes the model in the desired direction, one decision at a time. The aim is to make the reward as great as possible.

Therefore, in order to become better, the model must fail. Although this exploratory and random process is extremely time-consuming, it is nevertheless the best way to stimulate the model's 'creativity'. After all, an RL model can take into account thousands of alternative scenarios and thus propose far-reaching solutions that humans simply cannot devise.

Improving sales margins

So where does reinforcement learning provide real added value? A good example is dynamic pricing. Here, the price of products is continuously adjusted in order to optimise the sales margins. A simple algorithm can optimise one objective, such as demand, but not the margin, because that depends on other parameters such as volume and price. The challenge thus is to combine both in the equation and this can be done using reinforcement learning. RL models not only think about demand and cost, but also continuously analyse market parameters such as seasons, sales, advertising campaigns, etc.

These more in-depth analyses lead to much higher sales margins. In addition, the manual adjustment of prices is also fully automated, which avoids human errors and enables retailers to react much faster to fluctuating markets.

Opportunities in business and policy making

It could be said that it is seemingly strange that online retailers do not flock to reinforcement learning. The online data of the competition makes it easy for them to compare prices with AI, just as easy as it is for potential customers to find other places in a couple of clicks.

On a broader scale, the business world remains relatively unaware of the power of reinforcement learning, despite its many perspectives. In the hot topic of policy making during these times of Covid-19, you could measure the level of social support for certain measures such as lockdowns or curfew. There is still a lot of room for more data driven analyses and research and RL models could be one of the tracks to investigate.

Importance of the production phase

Awareness of the power of reinforcement learning can and must therefore grow dramatically. To achieve this, as many RL models as possible must reach the production phase. This is a goal we have always had in mind at Dataroots, and fortunately it is also starting to spread through the AI landscape. In the past, data scientists did not get much further than insights or models that ended up on the customer's shelf. Now, on the other hand, data scientists turn into engineers who get models past that production threshold. The industry is becoming more mature, the momentum is growing and reinforcement learning keeps pushing the boundaries of what is possible with AI.

Image credentials

Want to talk to our experts?

We're here for a chat!