Joint AI Seminar / Doctoral Speaking Skills Talk - Alexander Goldberg

— 1:00pm

Location:
In Person and Virtual - ET - ASA Conference Room, Gates Hillman 6115 and Zoom

Speaker:
ALEXANDER GOLDBERG, Ph.D. Student, Computer Science Department, Carnegie Mellon University
https://akgoldberg.github.io/research/

Scientific peer review plays a crucial role in maintaining the credibility of scholarly work. Concerns about its integrity and quality have motivated increasing transparency into the peer review process. In this talk, I will examine mechanisms for sharing more data about peer review while preserving reviewer anonymity. Our analysis of peer review highlights more general privacy-utility trade-offs that arise in sharing time series and graph-structured data. 

First, I will discuss “open peer review” systems (like OpenReview.net), where reviews are publicly posted while keeping reviewer identities anonymous. We show that temporal batching patterns in review activity create realistic risks of reviewer de-anonymization. We then propose a privacy-preserving mechanism that  adds random delays to review postings and prove that this mechanism optimally balances privacy risks and system latency under a formal privacy framework. 

Next, I will explore the challenge of sharing private data about reviewer-paper assignments to facilitate the evaluation of fraud detection algorithms. I will show that reporting even a single accuracy statistic on this dataset can enable a malicious third party to de-anonymize reviewers. Then, I will consider the use of techniques from differential privacy to mitigate these risks and outline potential directions for improving privacy-preserving data sharing in peer review. 

Presented as part of the Artificial Intelligence Seminar Series 

Presented in Partial Fulfillment of the CSD Speaking Skills Requirement 

In Person and Zoom Participation.  See announcement.

Add event to Google
Add event to iCal