Mitigating the Likelihood Paradox in Flow-based OOD Detection via Entropy Manipulation
This addresses a critical issue in OOD detection for deep generative models, offering an incremental improvement without retraining.
The paper tackles the likelihood paradox in flow-based out-of-distribution (OOD) detection by manipulating input entropy based on semantic similarity, showing that this increases the expected log-likelihood gap between in-distribution and OOD samples. It achieves consistent AUROC improvements over baselines on standard benchmarks.
Deep generative models that can tractably compute input likelihoods, including normalizing flows, often assign unexpectedly high likelihoods to out-of-distribution (OOD) inputs. We mitigate this likelihood paradox by manipulating input entropy based on semantic similarity, applying stronger perturbations to inputs that are less similar to an in-distribution memory bank. We provide a theoretical analysis showing that entropy control increases the expected log-likelihood gap between in-distribution and OOD samples in favor of the in-distribution, and we explain why the procedure works without any additional training of the density model. We then evaluate our method against likelihood-based OOD detectors on standard benchmarks and find consistent AUROC improvements over baselines, supporting our explanation.