Reinforcement Learning

Explanation of reinforcement learning using SmallTrain

SmallTrain can do many things. This section describes a reinforcement learning project using SmallTrain.

Reinforcement learning version of Small Train has not been released as an open source project (as of 2020.4)

Geek Guild has a high level of technology.

Geek Guild is a group of engineers with research experience and high engineering ability, You can develop a beginning state-of-the-art AI reinforcement learning.

Reinforcement learning is an advanced technology in artificial intelligence research and development in the field of prediction and simulation. There are various papers published in the academic field but in the business filed it is underdeveloped technology. The reason is that it requires a high level of technical skill to build a cutting edge reinforcement learning artificial intelligence.

Overview of Reinforcement Learning

  • Reinforcement learning is a learning method that builds a simulator (virtual environment) where agents (AI) learn, and makes trial and error in it.
  • Unlike supervised learning, there is no teacher that explicitly shows correct action output for state input
  • Learning through trial and error may discover better solutions than those obtained by human experts
  • All kinds of control methods are automatically acquired by the same agent, simply by setting how the reward is given by the designer according to the purpose.

Reinforcement Learning Projects

1. Logistic Order Automation

Goods for retailers are shipped from wholesalers’ warehouses to retailers in good time. It is necessary to keep it at a certain level because the shipping fee is reduced by finishing the shipment at one time, while the reduction in the number of shipments tends to increase the out-of-stock rate. Furthermore, you need to carry as much luggage as you can into one truck.

For complex tasks it is necessary to build state-of-the-art reinforcement learning models and improve accuracy.

A3C A3C is at the forefront of reinforcement learning. Go AI, which is famous for winning “Artificial Intelligence vs. World’s Strongest Player” in March 2016, used reinforcement learning called Q-Learning. Then we are using the most advanced reinforcement learning model that is three generations further.

2. Hotel Room Dynamic Pricing to Maximize Sales

We build a reinforcement learning model for hotel room prediction that maximizes sales In supervised learning where the past person’s pricing is the teacher, as the teacher data, it cannot be truly correct data. With changes in the environment around the hotel and changes in human behavior, we build a model that aims to maximize sales.