Towards a Principled Evaluation of Knowledge Editors
This work addresses the robustness and fairness of evaluation methods for knowledge editors, which is crucial for researchers and practitioners in AI to ensure reliable model editing, but it is incremental as it builds on existing datasets and highlights issues without proposing a new solution.
The paper tackled the problem of evaluating knowledge editors in machine learning by showing that different metrics, methodologies, and edit batch sizes can lead to inconsistent rankings of editors, and it revealed that string matching methods often produce false positives. The result included a demonstration of this effect on general language understanding tasks and a manual assessment exposing evaluation flaws.
Model editing has been gaining increasing attention over the past few years. For Knowledge Editing in particular, more challenging evaluation datasets have recently been released. These datasets use different methodologies to score the success of editors. Yet, it remains under-explored how robust these methodologies are and whether they unfairly favor some editors. Moreover, the disruptive impact of these editors on overall model capabilities remains a constant blind spot. We address both of these problems and show that choosing different metrics and evaluation methodologies as well as different edit batch sizes can lead to a different ranking of knowledge editors. Crucially we demonstrate this effect also on general language understanding tasks evaluated alongside the knowledge editing tasks. Further we include a manual assessment of the string matching based evaluation method for knowledge editing that is favored by recently released datasets, revealing a tendency to produce false positive matches.