Vulnerability of Text-Matching in ML/AI Conference Reviewer Assignments to Collusions

Explore how automated reviewer assignment systems in top-tier machine learning and AI conferences can be manipulated through text-matching vulnerabilities in this 15-minute conference presentation. Learn about the dual-factor assignment process that relies on reviewer bids and text similarity between reviewers' published work and submitted manuscripts, then discover how collusion rings can exploit the machine learning-based text-matching component even without bid manipulation. Examine specific security weaknesses in current assignment algorithms used at major ML/AI venues and understand how colluding parties can strategically game the system to ensure favorable reviewer assignments. Gain insights into the research findings that challenge the assumed security of text-based matching systems and review proposed solutions to strengthen the robustness of peer review processes against sophisticated manipulation attempts.