Special Colloquium
Alex Psomas
CMU
Theory and Practice of Fair Resource Allocation
Abstract:
The Internet and the vast increase in the availability of data have transformed algorithm design, as well as computer science in general. Designing algorithms that, given an input, quickly produce an output is no longer sufficient. In a myriad of modern applications, considerations like fairness and users’ incentives must be taken into account, complicating how success is defined and achieved. In this talk I’ll demonstrate how to tackle such issues in the context of food waste. I'll present a full stack of results on fair allocation, from theoretical results in formal models, all the way down to an improved allocation of food.
In the first part of the talk we study, from a purely theoretical angle, the fundamental problem of allocating a set of indivisible goods that arrive over time. Specifically, we will design algorithms that make allocation decisions in a way that is fair, under a formal definition of fairness. In the second part of the talk, we adopt and further develop an emerging paradigm called virtual democracy. Virtual democracy is an approach to automating decisions, by learning models of the preferences of individual people from data, and, at runtime, aggregating the predicted preferences of those people on the dilemma at hand. We will take the virtual democracy approach all the way to practice, in a collaboration with a Pittsburgh-based food bank, 412 Food Rescue, that provides on-demand food donation distribution services. I will present my work on designing and deploying an algorithm that automatically makes the decisions they most frequently face: given an incoming food donation, which recipient organization (such as a housing authority or food pantry) should receive it? I will also discuss challenges and solutions we faced in the data collection, learning and aggregation steps of virtual democracy, as well as this work’s implications for algorithmic fairness in general. I will conclude the talk by surveying some of my current and future research in algorithmic economics in general.
Colloquium tea to follow at 4pm in SEO 300.
Monday February 25, 2019 at 3:00 PM in 636 SEO