Understanding and Augmenting Human Behavior in Social Computing Systems

17 Apr
Tuesday, 04/17/2018 4:00pm to 5:00pm
Computer Science Building, Room 151

Abstract:  Social computing systems include, among others, social media and collaborative knowledge-based systems. In these systems, human behavior is entangled with system design. Przemyslaw conducts statistical studies of social processes happening in these systems and investigates how these processes could inform system design. For instance, we performed an experiment on social influence with 2200 participants, by creating a mirror proxy of YouTube. We find that the opinions of participants about videos are systematically influenced by the comments shown underneath these videos. He argues that the knowledge about social influence is important for designing fair and efficient rating subsystems. Furthermore, in the contemporary techno-social systems, human behavior is often augmented with machine-learning subsystems for recommendation and prediction. A flourishing line of research studies how to train machines that do not discriminate against race or gender, while maximizing their utility in the process of training on historical observations potentially tainted with discrimination. He proposes the first evaluation framework for such "fair" training methods. Additionally, he introduces the first method that prevents both direct discrimination and the inducement of any indirect discrimination. Our fair training method performs significantly better than the state-of-the-art methods in the proposed evaluation framework. This method is universally applicable under major statistical inference paradigms and has efficient implementations for specific models. We plan to apply the proposed fair and efficient rating subsystems and training methods in a privacy-preserving social computing system for open science, which he will introduce at the end of this presentation.

Bio:  Przemyslaw A. Grabowicz received his Ph.D. in Interdisciplinary Physics from the Institute for Cross-Disciplinary Physics and Complex Systems (2014, cum laude) and M.Sci. in Applied Physics from Warsaw University of Technology (2008, cum laude). Since October 2013, Przemyslaw is a Postdoctoral Fellow at the Max Planck Institute for Software Systems, where he is a member of the social computing group headed by Prof. Krishna P. Gummadi. Przemyslaw has an interdisciplinary training and research experience. He has performed long research stays in computer science, math, physics, and industry research labs. Recently, he has been awarded a prestigious grant from the Volkswagen Foundation for a computational social science project on agenda setting. 


A reception for attendees will be held at 3:30 p.m. in CS 150

Faculty Host