Wouldn’t it be amazing if machines could be able to think like us and learn some stuff like cooking, playing some sports, responding to phone calls all by themselves? Well we are actually heading in that direction and future is probably not that far. This particular field of research comes under the umbrella of Reinforcement Learning (RL). RL is amazing because using this approach we can actually program some self learning behaviors into machines and let them learn anything by trying and gaining experience. In this article to give an idea about how we can program such “self learning machines”, I am going to explain how we can train some matchboxes to play Tic-tac-toe with human level performance just by playing a few games with them. Moreover, we will extend this idea further and write a tic-tac-toe program in python that learns how to play the game by playing thousands of games with itself within a few seconds.
Have you heard of “sleep sort” algorithm? I recently came across it while searching for the most out of the box algorithms on the internet. So I decided to implement this in c plus plus and try it by myself. In this tutorial, we will implement this algorithm in c plus plus. While doing so, we’ll learn:
Bayes’ rule is amazing. Ever since I came across Bayes’ rule in statistics I have been in love with this simple yet elegant formula in statistics which has application in wide area of research such as AI, health care, economics and what else. Although it looks very simple, in many cases it becomes very complicated to compute it in closed form. This complication arises due to the denominator of the formula which is the marginal probability distribution of the observations. This term is sometimes very hard (or virtually impossible) to compute analytically. To bypass this complexity, several approaches has been proposed to approximate the Bayesian posterior such as Variational inference, Markov Chain Monte Carlo (MCMC) etc . In this informal tutorial we are going to see how we can approximate the Bayes rule using MCMC.
Finally, it’s my pleasure to announce that our paper with the title “Multi-objective Model-based Policy Search for Data-efficient Learning with Sparse Rewards” has been accepted in Conference on Robot Learning (CoRL) - 2018 which will be held on October 29th-31st, 2018, in Zürich, Switzerland. This time the acceptance rate in CoRL was 31%. Only 75 out of 237 papers have been accepted.