Strengthen Out-of-Distribution Detection Capability with Progressive Self-Knowledge Distillation

Out-of-distribution (OOD) detection aims to ensure AI system reliability by rejecting inputs outside the training distribution. Recent work shows that memorizing atypical samples during later stages of training can hurt OOD detection, while strategies for forgetting them show promising improvements. However, directly forgetting atypical samples sacrifices ID generalization and limits the model's OOD detection capability. To address this issue, we propose Progressive Self-Knowledge Distillation (PSKD) framework, which strengthens the OOD detection capability by leveraging self-provided uncertainty-embedded targets. Specifically, PSKD adaptively selects a self-teacher model from the training history using pseudo-outliers, facilitating the learning of uncertainty knowledge via multi-level distillation applied to features and responses. As a result, PSKD achieves better ID generalization and uncertainty estimation for OOD detection. Moreover, PSKD is orthogonal to most existing methods and can be integrated as a plugin to collaborate with them. Experimental results from multiple OOD scenarios verify the effectiveness and general applicability of PSKD.

Deep learning models can make mistakes when they see out-of-distribution (OOD) data—inputs that fall outside the training distribution—but still act very confidently about those wrong answers. This overconfidence raises safety and reliability concerns. Recently, researchers have found that deep learning models tend to memorize atypical samples late in training, which makes them overly confident about these atypical samples and harder to detect when something is OOD. Some methods try to make the model forget these atypical samples, but that hurts its performance on regular tasks.In this paper, we develop a new method called Progressive Self-Knowledge Distillation (PSKD) to help models better detect OOD inputs through self-teaching. PSKD allows the model not only to learn the correct answers but also to understand how confident it should be in those answers. As a result, PSKD improves both the model’s performance on regular tasks and its ability to identify OOD data. Additionally, PSKD works well with many existing methods, making it useful for improving AI safety and reliability in various situations.