Auditing Visualizations: Transparency Methods Struggle to Detect Anomalous Behavior
This work addresses the reliability of transparency methods for machine learning practitioners, but it is incremental as it primarily evaluates existing methods rather than proposing a new solution.
The paper tackled the problem of whether model visualizations can reliably detect anomalous behaviors like backdoors or overregularization, finding that existing methods struggle with subtle anomalies and fail to identify inducing inputs, revealing significant limitations.
Model visualizations provide information that outputs alone might miss. But can we trust that model visualizations reflect model behavior? For instance, can they diagnose abnormal behavior such as planted backdoors or overregularization? To evaluate visualization methods, we test whether they assign different visualizations to anomalously trained models and normal models. We find that while existing methods can detect models with starkly anomalous behavior, they struggle to identify more subtle anomalies. Moreover, they often fail to recognize the inputs that induce anomalous behavior, e.g. images containing a spurious cue. These results reveal blind spots and limitations of some popular model visualizations. By introducing a novel evaluation framework for visualizations, our work paves the way for developing more reliable model transparency methods in the future.