Over the past 5 years Machine Learning techniques have dramatically reshaped the landscape of traditional, rule based software. This trend emerged out of decades of academic research but is now rapidly finding its way to an ever increasing number of businesses.
And yet, most of the ML systems in use today are what we could call “Static Inference Machines”: take input sample X, produce output Y and repeat. Standard examples of this framework include image classification, machine translation, music recommendation and many more. Additionally, almost all practical systems to date are trained with supervised learning, requiring large volumes of labelled training data: a major drawback in many practical settings.
As a response to these inherent restrictions, this talk will focus on newly emerging paradigms in intelligent software design such as Reinforcement Learning that enable systems to learn from sparse reward signals in dynamic environments.
From a global theoretical overview of the problem setting to some fascinating practical applications, this talk will try to shed some light on the future of Machine Learning.
About the speaker:
Xander Steenbrugge is researcher at ML6, researcher on Reinforcement Learning. He's also the youtube channel Arxiv Insights: https://www.youtube.com/channel/UCNIkB2IeJ-6AmZv7bQ1oBYg