Reference-Specific Unlearning Metrics Can Hide the Truth: A Reality Check
This addresses a critical evaluation gap in machine unlearning for generative models, exposing systematic blind spots that could mislead development, though it is incremental in improving assessment rather than unlearning itself.
The paper tackles the problem that current unlearning metrics for generative models rely on reference-specific evaluations, which can hide retained unwanted knowledge, and proposes FADE, a new metric that measures distributional similarity, revealing failures in existing methods on benchmarks like TOFU and UnlearnCanvas.
Current unlearning metrics for generative models evaluate success based on reference responses or classifier outputs rather than assessing the core objective: whether the unlearned model behaves indistinguishably from a model that never saw the unwanted data. This reference-specific approach creates systematic blind spots, allowing models to appear successful while retaining unwanted knowledge accessible through alternative prompts or attacks. We address these limitations by proposing Functional Alignment for Distributional Equivalence (FADE), a novel metric that measures distributional similarity between unlearned and reference models by comparing bidirectional likelihood assignments over generated samples. Unlike existing approaches that rely on predetermined references, FADE captures functional alignment across the entire output distribution, providing a principled assessment of genuine unlearning. Our experiments on the TOFU benchmark for LLM unlearning and the UnlearnCanvas benchmark for text-to-image diffusion model unlearning reveal that methods achieving near-optimal scores on traditional metrics fail to achieve distributional equivalence, with many becoming more distant from the gold standard than before unlearning. These findings expose fundamental gaps in current evaluation practices and demonstrate that FADE provides a more robust foundation for developing and assessing truly effective unlearning methods.