Introduction

Due to Deep Learning, the technology to learn from large data set has evolved exponentially and now, we are able to recognize certain objects and increase translation accuracy, especially for image and voice recognition purpose. There have even been reports that these models outperform specialists in certain fields. However, when we look back, we realize that humans can learn without input data. In machine learning, there is a machine learning field called “Reinforcement Learning”, which bases its algorithm on finding the optimal action using the intuitive human behavior of trial and error.

Reinforcement learning is a method where the agent learns the optimal input scheme from rewards only. Applying pictures and sensor data to reinforcement learning was a difficult task, due to the finite state the algorithm can handle, but due to appliance of deep learning, more difficult problems can be handled where problems use image or sensory data.

Recently, information about reinforcement learning on the internet has increased, but without sufficient knowledge and enough experience of applying to real world problems, research becomes time consuming, and difficulty of applying reinforcement learning to real world problem arises. To overcome this problem, we thought of providing “ReNomRL”, an API that can help people understand this field more clearly.