Alekh Agarwal explains why interactive learning systems that go beyond the routine train/test paradigm of supervised machine learning are essential to the development of AI agents. Along the way, Alekh outlines the novel challenges that arise at both the systems and learning side of things in designing and implementing such systems.
Alekh begins by discussing addresses a class of interactive learning problems called contextual bandits while exploring a system recently developed at Microsoft. The system is available for public use via Azure.
Alekh then looks forward to the challenges underlying the implementation of systems for more general reinforcement learning in a stateful world, discussing the difficulties of prototyping and evaluating such systems rapidly, at scale, and across a diverse set of problems unlike classical supervised learning before describing Malmo, a novel open source AI experimentation framework built on top of the game Minecraft. Alekh explains how this platform provides a flexible environment for evaluating AI agents across a diverse array of tasks.
Alekh Agarwal is a researcher at Microsoft Research New York City working on machine learning. His research spans several areas, including online learning and optimization and learning with partial feedback, which routinely arises in interactive machine learning and reinforcement learning problems. Alekh has won several awards, including a best paper award at NIPS 2015.
©2016, O'Reilly Media, Inc. • (800) 889-8969 or (707) 827-7019 • Monday-Friday 7:30am-5pm PT • All trademarks and registered trademarks appearing on oreilly.com are the property of their respective owners. • firstname.lastname@example.org