Machine Learning has revolutionized our everyday lives. All of us rely on complex algorithms that learn our preferences over products, services, and even news articles. The deployed algorithms are always strongly vetted, with theoretical performance guarantees being provided against both stochastic and adversarial attackers. But what if these models of attacks fail to fully capture the goals and desires of real-life attackers? Think about the spread of Fake News, for example. The agents spreading Fake News are agents trying to take advantage of the way the online news serving algorithms work; bots make sure to appropriately bump a Fake News article, so that more users are exposed to it. These carefully crafted strategic manipulations should be dealt with in a different way than standard adversarial attacks, whose goal is to destroy the news algorithms. Similar strategic incentives arise in many scenarios, such as classification algorithms for evaluation, algorithms for Stackelberg Security Games and many more.
Such problems, situated at the interface of Algorithmic Game Theory and Machine Learning, present a unique challenge, but also a unique opportunity to modern researchers. The goal of this tutorial is to bring the vibrant research agenda, which bridges between Economics and Machine Learning, to the forefront of the EC community, by presenting a number of results published in Machine Learning and Algorithmic Game Theory venues.
|Pre-recording Part I||Wednesday, June 17||1:30-2:15pm, 2:30-3:15pm Eastern|
|Exercise Session||Wednesday, June 17||3:30-4:30pm Eastern|
|Pre-recording Part II||Thursday, June 18||1:30-2:15pm, 2:30-3:15pm Eastern|