SEA: How to optimize for a wide variety of users?

This SEA will be all about optimizing ML systems for a wide variety of users. We have two amazing speakers lined up: Preethi Lahoti from the Max Planck Institute for Informatics and Rishabh Mehrotra from Spotify research.

Note that this meetup is not at the usual time slot! This SEA will take place from 14:00 - 15:30 CET.

*** IMPORTANT: Make sure to (1) attend the meetup on the meetup page and (2) ensure you receive emails from Meetup. Shortly before the event we will send you the Zoom link and password to attend, as well as the info you need to log in via the browser (if your organisation does not allow you to install Zoom). You will only receive this if you have done both these steps. ***

** 14:00 - 14:30 CET - Preethi Lahoti, Max Planck Institute for Informatics **

Title: Operationalizing Fairness in Practice: Challenges and Approaches

Abstract: As machine learning is increasingly used for decision making in scenarios that impact humans, there is a growing awareness of its potential for unfairness. A flurry of recent work has focused on proposing formal notions of fairness in machine learning, as well as approaches to mitigate unfairness. However, there is a growing disconnect between ML fairness literature and the needs of the ML practitioners to operationalize fairness in practice. For instance, most fairness approaches assume that protected features such as race are accessible, and rely upon them to mitigate unfairness. In practice, however, factors like privacy and regulation often prohibit ML models from collecting and/or using protected features. Similarly, most fairness approaches focus on distributive fairness, which reduces fairness to a group statistic. However, automated decisions are made at an individual level, and can often adversarially impact individual members of the society irrespective of the group statistic.
In this talk, after a brief introduction to fairness in machine learning, I will formalize the aforementioned challenges in operationalizing fairness in practice, and outline approaches to address them. I will conclude by embedding these works in a broader context of fairness and machine learning and highlighting directions for future work.

** 14:30 - 15:00 CET - Rishabh Mehrotra, Spotify Research **

Title: Recommendation aspects of Multi-stakeholder Marketplace Optimization

Abstract: Multi-sided marketplaces facilitate efficient interactions between multiple stakeholders, including e.g. buyers and retailers (Amazon), guests and hosts (AirBnb), riders and drivers (Uber), and listeners and artists (Spotify). Recommender systems powering online multi-stakeholder platforms often face the challenge of jointly optimizing multiple objectives, in an attempt to efficiently match suppliers and consumers. In this talk, I will touch upon some recent advances in designing recommendation systems that power such platforms, including (1) multi-objective bandits, (2) user & content aware reward modeling, (3) role of consumption & supplier diversity and (4) interplay between stakeholder objectives. I will discuss insights from large scale experiments, and highlight important research directions.

After the two talks we leave the Zoom call open for another half an hour, for any remaining questions. Last time this resulted in a nice discussion!