The Signal in the Noise: OOD Detection Through Goodness-of-Fit Testing in Factorised Latent Spaces
arXiv:2605.22496v1 Announce Type: new Abstract: Deep generative models offer a natural foundation for out-of-distribution (OOD) detection, yet prior work has shown that their assigned likelihoods are notoriously unreliable indicators for in- vs out-of-distribution data. In this paper, we address this problem by leveraging the diffeomorphic and mass-preserving properties of continuous normalising flows. Our analysis shows that OOD samples are mapped to noise samples that are highly atypical under the noise prior in ways not captured by the likelihood. Based on this observation, we propose a new
The paper addresses a known limitation in deep generative models for OOD detection, applying continuous normalising flows to improve reliability.
Improved out-of-distribution (OOD) detection is critical for the safety, robustness, and trustworthiness of AI systems deployed in real-world scenarios.
This research enhances the ability of AI models to identify novel or anomalous inputs, moving beyond unreliable likelihood-based methods.
- · AI safety researchers
- · Generative AI developers
- · Autonomous system manufacturers
- · Cybersecurity sector
- · AI systems with poor OOD detection
- · Legacy anomaly detection methods
AI systems become more capable of identifying inputs that deviate from their training data, leading to safer deployment.
Increased trust in AI applications, particularly in high-stakes environments like autonomous vehicles or medical diagnostics, accelerates adoption.
Robust OOD detection becomes a standard and expected feature for certification of advanced AI, creating a new competitive dimension for model providers.
This signal links to a primary source. Continuum Brief monitors and indexes it as part of the live intelligence stream — we do not republish source content.
Read at arXiv cs.LG