This is Shakil from Product Engineering Department. This time I wanted to talk a bit about Markov Decision Processes which forms the basis of Reinforcement Learning.
What are MDPs (Markov Decision Processes)?
MDP provides a mathematical toolset to model Problems involving decision making in sequences where what the future state is going to be depends only on the actions taken from the present state.
Why are MDPs interesting?
MDPs can be used to model and solve a wide range of problems, from Self-Driving Cars, to Recommendation Systems to Complex Video Games like DOTA 2.
Open AI 5 by DeepMind defeated human world Champions of DOTA 2 in 2019 by using MDPs as one of the tools to model and master the game.
続きを読む