The ICML 2023 Ranking Experiment: Examining Author Self-Assessment in ML/AI Peer Review

2024-08-24

AI Summary

This study addresses score bias and low inter-rater reliability in machine learning conference peer review. For the first time, it conducts a large-scale empirical study at ICML 2023, eliciting authors’ self-ranked orderings of their own multiple submissions by quality. We propose the Isotonic Mechanism—a calibration method grounded in isotonic regression—that maps raw reviewer scores to calibrated scores consistent with authors’ ordinal preferences, ensuring interpretability and minimal intervention risk. Experiments show that calibrated scores significantly outperform raw scores in predicting “ground-truth expected review scores,” with substantial reductions in MSE and MAE. Ordinal consistency analysis confirms the reliability of author-provided rankings. The mechanism has been operationally deployed to support area chair oversight, award selection, and emergency reviewer assignment—establishing a practical, deployable paradigm for enhancing human-AI collaborative review quality.

Technology Category

• Machine Learning: Calibration & Uncertainty Quantification
• Humans and AI: Planning and Decision Support for Human-Machine Teams
• Multiagent Systems: Mechanism Design

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• User Modeling, Personalization and Recommendation: Fairness-aware retrieval and ranking
• Economics, Online Markets and Human Computation: LLM based quality controls for crowd work

Abstract

We conducted an experiment during the review process of the 2023 International Conference on Machine Learning (ICML), asking authors with multiple submissions to rank their papers based on perceived quality. In total, we received 1,342 rankings, each from a different author, covering 2,592 submissions. In this paper, we present an empirical analysis of how author-provided rankings could be leveraged to improve peer review processes at machine learning conferences. We focus on the Isotonic Mechanism, which calibrates raw review scores using the author-provided rankings. Our analysis shows that these ranking-calibrated scores outperform the raw review scores in estimating the ground truth ``expected review scores'' in terms of both squared and absolute error metrics. Furthermore, we propose several cautious, low-risk applications of the Isotonic Mechanism and author-provided rankings in peer review, including supporting senior area chairs in overseeing area chairs' recommendations, assisting in the selection of paper awards, and guiding the recruitment of emergency reviewers.

Problem

• Examining author self-assessment in ML/AI peer review
• Improving peer review using author-provided rankings
• Calibrating review scores with the Isotonic Mechanism

Innovation

• Used Isotonic Mechanism for score calibration
• Leveraged author rankings to improve peer review
• Applied rankings for awards and reviewer selection