Joint Seminar by Andrew Heeszel and Kiljae Lee
Abstract:
In this talk, we show a law of large numbers and central limit theorem for the edge speeds of the Boundary Modified Contact Process (BMCP). The BMCP models an epidemic spreading across the integer lattice with two infection parameters. Starting from a finite infected set, each edge transmits the infection at rate except for the rightmost and leftmost edges incident to infected vertices. We study the model when the interior infection rate is at the critical infection rate of the contact process and cannot sustain the infection. Adding a boost to the exterior infection rates causes the BMCP to no longer be an attractive particle system, requiring new tools in its study. We also show the likelihood of the infection dying out after a long but finite time scales at the stretched exponential rate, in contrast with the supercritical contact process.
Talk 2 — Kiljae Lee (PhD Candidate, The Ohio State University)
Title: Robust Shapley-Based Data Valuation under Strategic Manipulation
Abstract:
Shapley-based data valuation has become a useful framework for evaluating data contributions, but modern data markets raise challenges that classical formulations do not fully address. In particular, valuation rules should remain reliable when participants can strategically repackage their data or when some data contributions depend on others through copying, reuse, or augmentation. In this talk, I present two recent approaches that address these issues. Faithful Group Shapley Value (FGSV) provides a group data valuation rule that is robust to shell-company style manipulations by ensuring that a group’s value does not depend on how other participants partition their data. On the other hand, Priority-Aware Shapley Value (PASV) extends Shapley-based valuation to settings with precedence constraints and soft priorities, enabling lineage-aware credit assignment that reduces overvaluation of copied or harmful data. Together, these works aim to make data valuation more robust, interpretable, and aligned with realistic data marketplace incentives.