![[2602.23360] Model Agreement via Anchoring](https://arxiv.org/static/browse/0.3.4/images/arxiv-logo-fb.png){kind=logo}
[2602.23360] Model Agreement via Anchoring
Summary
The paper presents a method for reducing model disagreement in machine learning by using an anchoring technique, demonstrating its effectiveness across various algorithms.
Why It Matters
Model disagreement can lead to inconsistent predictions in machine learning systems. This research addresses a critical issue by providing a framework to minimize disagreement, which is essential for improving model reliability and performance in real-world applications.
Key Takeaways
- Introduces a technique to reduce model disagreement using anchoring.
- Proves disagreement bounds for multiple machine learning algorithms.
- Demonstrates applicability in one-dimensional and multi-dimensional regression.
- Highlights the importance of model coordination in training processes.
- Provides a foundation for future research in model agreement methodologies.
Computer Science > Machine Learning arXiv:2602.23360 (cs) [Submitted on 26 Feb 2026] Title:Model Agreement via Anchoring Authors:Eric Eaton, Surbhi Goel, Marcel Hussing, Michael Kearns, Aaron Roth, Sikata Bela Sengupta, Jessica Sorrell View a PDF of the paper titled Model Agreement via Anchoring, by Eric Eaton and 6 other authors View PDF HTML (experimental) Abstract:Numerous lines of aim to control $\textit{model disagreement}$ -- the extent to which two machine learning models disagree in their predictions. We adopt a simple and standard notion of model disagreement in real-valued prediction problems, namely the expected squared difference in predictions between two models trained on independent samples, without any coordination of the training processes. We would like to be able to drive disagreement to zero with some natural parameter(s) of the training procedure using analyses that can be applied to existing training methodologies. We develop a simple general technique for proving bounds on independent model disagreement based on $\textit{anchoring}$ to the average of two models within the analysis. We then apply this technique to prove disagreement bounds for four commonly used machine learning algorithms: (1) stacked aggregation over an arbitrary model class (where disagreement is driven to 0 with the number of models $k$ being stacked) (2) gradient boosting (where disagreement is driven to 0 with the number of iterations $k$) (3) neural network training with archit...
