Edit at your own risk: evaluating the robustness of edited models to distribution shifts

TL;DR

This paper investigates how model editing impacts the robustness of models to distribution shifts, revealing that edits often reduce robustness and proposing a new editing algorithm to balance accuracy and robustness.

Contribution

It introduces a novel model editing algorithm, 1-layer interpolation (1-LI), that balances editing accuracy with model robustness, addressing a key gap in current research.

Findings

Model edits tend to decrease overall robustness.

The degree of robustness degradation depends on the editing method and layers used.

The proposed 1-LI algorithm helps balance editing accuracy and robustness.

Abstract

The current trend toward ever-larger models makes standard retraining procedures an ever-more expensive burden. For this reason, there is growing interest in model editing, which enables computationally inexpensive, interpretable, post-hoc model modifications. While many model editing techniques are promising, research on the properties of edited models is largely limited to evaluation of validation accuracy. The robustness of edited models is an important and yet mostly unexplored topic. In this paper, we employ recently developed techniques from the field of deep learning robustness to investigate both how model editing affects the general robustness of a model, as well as the robustness of the specific behavior targeted by the edit. We find that edits tend to reduce general robustness, but that the degree of degradation depends on the editing algorithm and layers chosen. Motivated by these observations we introduce a new model editing algorithm, 1-layer interpolation (1-LI), which uses weight-space interpolation to navigate the trade-off between editing task accuracy and general robustness.