{"id":708913,"date":"2020-12-07T07:59:52","date_gmt":"2020-12-07T15:59:52","guid":{"rendered":"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/?p=708913"},"modified":"2020-12-07T20:39:02","modified_gmt":"2020-12-08T04:39:02","slug":"neurips-2020-moving-toward-real-world-reinforcement-learning-via-batch-rl-strategic-exploration-and-representation-learning","status":"publish","type":"post","link":"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/blog\/neurips-2020-moving-toward-real-world-reinforcement-learning-via-batch-rl-strategic-exploration-and-representation-learning\/","title":{"rendered":"NeurIPS 2020: Moving toward real-world reinforcement learning via batch RL, strategic exploration, and representation learning"},"content":{"rendered":"\n
\"diagram,<\/figure>\n\n\n\n

As human beings, we encounter unfamiliar situations all the time\u2014learning to drive, living on our own for the first time, starting a new job. And while we can anticipate what to expect based on what others have told us or what we\u2019ve picked up from books and depictions in movies and TV, it isn\u2019t until we\u2019re behind the wheel of a car, maintaining an apartment, or doing a job in a workplace that we\u2019re able to take advantage of one of the most important means of learning: by trying. We make deliberate decisions, see how they pan out, then make more choices and take note of those results, becoming\u2014we hope\u2014better drivers, renters and workers in the process. We learn by interacting with our environments.<\/p>\n\n\n\n

\u201cHumans have an intuitive understanding of physics, and it\u2019s because when we\u2019re kids, we push things off of tables and stuff like that,\u201d says Principal Researcher Akshay Krishnamurthy<\/a>. \u201cBut if you only watch videos of things falling off tables, you will not actually know about this intuitive gravity business. So our ability to do experimentation in the world is very, very important for us to generalize.\u201d <\/p><\/blockquote>\n\n\n\n

For our AI to improve in the world in which we operate, it would stand to reason that our technology be able to do the same. To learn not just from the data it\u2019s been given, as has largely been the approach in machine learning, but to also learn to figure out what additional data it needs to get better. <\/p>\n\n\n\n

\u201cWe want AIs to make decisions, and reinforcement learning is the study of how to make decisions,\u201d says Krishnamurthy.   <\/p>\n\n\n\n

\n\t
\n\t\t
\n\t\t\t\t\t\t\t\n\t\t\t\t\t\"illustrated\t\t\t\t<\/a>\n\t\t\t\t\t\t\tEVENT<\/span>\n\t\t\tMicrosoft at NeurIPS 2020 <\/span> <\/span><\/a>\t\t\t\t\t\t\t

Check out Microsoft at NeurIPS 2020, including all of our NeurIPS publications, the Microsoft session schedule, and open career opportunities<\/p>\n\t\t\t\t\t<\/div>\n\t<\/article>\n<\/div>\n\n\n\n

Krishnamurthy is a member of the reinforcement learning group at the Microsoft Research lab in New York City, one of several teams helping to steer the course of reinforcement learning at Microsoft. There are also dedicated groups in Redmond, Washington; Montreal; Cambridge, United Kingdom; and Asia; and they\u2019re working toward a collective goal: RL for the real world<\/a>. They\u2019ve seen their efforts pay off. The teams have translated foundational research into the award-winning Azure Personalizer (opens in new tab)<\/span><\/a>, a reinforcement learning system that helps customers build applications that become increasingly customized to the user, which has been successfully deployed in many Microsoft products, such as Xbox.<\/p>\n\n\n\n

While reinforcement learning has been around almost as long as machine learning, there’s still much to explore and understand to support long-term progress with real-world implications and wide applicability, as underscored by the 17<\/em> RL-related papers being presented by Microsoft researchers at the 34th Conference on Neural Information Processing Systems (NeurIPS 2020)<\/a>. Here, we explore a selection of the work through the lens of three areas:<\/p>\n\n\n\n