Date: Wednesday, February 15, 2023
9:00 am – 10:00 am Pacific Time
12:00 pm – 1:00 pm Eastern Time
Location: Weekly Seminar, Zoom
Title: Individual Fairness in Online Classification
The bulk of the literature in algorithmic fairness asks for fairness on a group or an aggregate level. From an individual’s standpoint, however, such guarantees promise very little – they only bind over averages over many people. On the contrary, the notion of individual fairness, proposed by the seminal work of Dwork et al. (2011), asks for models that treat similar individuals similarly. In their formulation, this is a Lipschitz constraint imposed on a randomized classifier, and who is “similar” is defined by a task-specific similarity metric. A major challenge with using the framework of individual fairness is the unavailability of similarity metrics in many domains.
Additionally, most of the literature in algorithmic fairness operates under statistical assumptions, where data is assumed to be generated in an i.i.d. fashion from some (unknown) fixed distribution. However, as problem domains where fairness is a major concern generally involve decision making over human individuals, such assumptions often cease to hold in practice, due to, e.g.: (i) individuals’ strategic behavior, (ii) distribution shifts over time, (iii) adaptivity to previous decisions. Things are further complicated by the one-sided feedback structure that is prevalent in most of these problem domains – where label information is available only for positively predicted individuals.
In this talk, I will discuss recent endeavors towards the design of practical algorithms for individual fairness in online settings, while attempting to minimize surrounding assumptions: regarding the availability or the parametric form of the similarity metric, the observable feedback for made decisions, and the data generation process.
Talk based on joint work with Christopher Jung, Aaron Roth, and Steven Wu.
Yahav Bechavod is a Postdoctoral Researcher at the School of Engineering and Applied Science at the University of Pennsylvania, working with Prof. Aaron Roth. Prior to joining Penn, He was a PhD student at the School of Computer Science and Engineering at the Hebrew University of Jerusalem and an Apple Scholar in AI/ML. His research interests are primarily in algorithms, machine learning, and game theory, and specifically in the areas of fairness in machine learning, online learning, and learning in the presence of strategic behavior. He is the recipient of several awards and fellowships, including the Israeli Council for Higher Education Postdoctoral Fellowship, the Apple Scholars in AI/ML PhD Fellowship, and the Charles Clore Foundation PhD Fellowship.