Reinforcement learning

How to Teach Computers

 

This article from Karthik Rajagopalan’s company blog introduces a type of machine learning that may provide solutions for some traditional optimization problems such as inventory optimization and supply chain optimization.

The field of artificial intelligence is slowly beginning to permeate our lives. Computers and other machines are being endowed with intelligence through a process called machine learning. One particular type of machine learning that is of interest to us here in Paramis Digital is Reinforcement Learning (RL), which has been quite successful in a limited number of applications like teaching computers how to play video games and teaching robots how to perform certain activities.

Supervised learning uses data with input and labeled output and learns the relationship between them. Unsupervised learning, on the other hand, works on unlabeled data and is generally used for pattern detection. It is not uncommon to find applications that use a combination of the two methods. In reinforcement learning an agent, say a robot, learns to perform activities in an environment to accomplish a specified goal. The agent learns by performing various activities, collecting feedback from the environment in response to the activities and evaluating the feedback. The key part of this learning method is the reward mechanism, a mathematical construct, which rewards the agent for performing activities that will move it towards and eventually accomplish the goal. The agent is also punished for performing activities that detract it from accomplishing its goals.

 

Read the full article, Reinforcement Learning, on the Paramis Digital website.