Surrogate model trust varies near decision boundaries and across trainings
"Where Can I Trust You?": Boundary-Aware Evaluation of Surrogate Fidelity
Machine Learning
Summary
When a simple model learns from a complex one, it is usually judged on how often it matches the complex model's answers. This work shows that the simple model matches less near the complex model’s decision edges, where choices are less clear. The authors also found that checking these boundary areas helps predict how well the simple model will agree with the complex one in new situations. Lastly, even similar complex models may disagree on where these decision boundaries lie, making them unstable points for evaluation.
What this means in practice
- •For machine learning engineers: Adjust the evaluation of surrogate models to focus on accuracy near decision boundaries for better understanding of model trustworthiness.
- •For autonomous system developers: Use boundary-aware fidelity checks to improve surrogate model reliability when deploying complex models in real-world systems.
Authors
Jackson Eshbaugh
Abstract
Surrogate models are commonly evaluated by how often they agree with their teacher model over an evaluation set. Local variation in this agreement is well known, but its structure and consequences are less clear. We ask whether disagreement is systematically concentrated near the teacher's decision boundary and whether retaining that structure provides information beyond a single global score. Across several datasets and surrogate model classes, we find substantially lower fidelity near teacher decision boundaries under two different methods of identifying near-boundary examples. Moreover, conditioning agreement on confidence-defined regions improves prediction of teacher--surrogate agreement when evaluation-set composition changes, relative to the global score alone. Yet surrogates that agree equally well with the teacher both globally and near the decision boundary can respond very differently to changes selected using the surrogate itself. Finally, we show that independently trained deep teachers can agree on most predictions while identifying different examples as lying near their decision boundaries, complicating the use of those boundaries as stable reference regions for evaluating surrogates. Together, these results show that surrogate fidelity depends not only on how often a surrogate agrees with its teacher, but also on where that agreement holds and, for deep models, how stable the teacher's decision boundary is across training runs.