Shahrzad Haddadan
Rutgers University
Optimally Improving Cooperative Learning in a Social Setting
Abstract
Many important economic and social decisions, beliefs, and attitudes arise from a learning process that depends on the exchange of imperfect information across a potentially large and complex social network involving heterogenous and self-interested agents. For example, an individual may seek to determine the veracity of a news story or image of consequence encountered on social media. Towards this end, they may interact with and gain information from other social media users or AI tools – search algorithms, AI-generated social media content, or large-language models.
In this work, we consider a cooperative learning scenario where a collection of networked agents with individually owned classifiers dynamically update their predictions, for the same classification task, through communication or observations of each other’s predictions. Clearly if highly influential vertices use erroneous classifiers, there will be a negative effect on the accuracy of all the agents in the network. This leads to the following question: how can we optimally help a few agents improve their prediction so as maximize the overall accuracy in the entire network. How can we optimally assist a few agents in improving their predictions to maximize the overall accuracy of the entire network? To address this, we consider both an aggregate and an egalitarian objective function and offer algorithmic solutions for achieving optimal improvement across the network.
Joint work with Cheng Xin and Jie Gao.
About
Shahrzad Haddadan is a theoretical computer scientist whose interest is primarily in the mathematical analysis of massive and complex data. She received her Ph.D. in Computer Science from Dartmouth College, privileged to be supervised by Prof. Peter Winkler.
After that, Haddadan spent some wonderful years as a postdoctoral researcher working under the supervision of Prof. Flavio Chierichetti at La Sapienza, University of Rome I, and later under the supervision of Prof. Eli Upfal at Brown University. Between these two terms, due to an unexpected delay in the processing of her visa application, she spent five months at Max-Planck Institut für Informatik as a guest researcher.
Haddadan's research program concentrates on the mathematical modeling of challenging situations arising in the analysis of big data by the usage of combinatorial tools; the goal of such mathematical modeling is the development of rigorous algorithms in complex situations, thus bridging the gap between theory and practice. She publishes her works in internationally renowned conferences and journals of Theoretical Computer Science, Machine Learning, Data Mining, and Graph Mining.