DynaSubVAE: Adaptive Subgrouping for Scalable and Robust OOD Detection

2025-06-11

AI Summary

Real-world observational data often contain dynamically evolving heterogeneous subpopulations; conventional models erroneously classify minority subgroups as out-of-distribution (OOD) samples—due to neglecting them—leading to inaccurate OOD detection and biased predictions. This paper proposes a dynamic latent subgroup modeling framework that shifts OOD detection from passive recognition to active modeling of emerging patterns. Built upon a variational autoencoder, the method integrates a GMM-inspired nonparametric similarity clustering with an online-evolving latent-space mechanism, jointly optimizing representation learning and OOD detection. It requires no pre-specified number of subgroups and enables real-time discovery and adaptation to newly emerging substructures. Evaluated on near-OOD, far-OOD, and class-level OOD benchmarks, our approach achieves state-of-the-art performance—particularly under class-missing scenarios—where it significantly outperforms baselines including GMM and KMeans++ in both OOD detection accuracy and regret-aware precision.

Technology Category

• Data Mining & Knowledge Management: Anomaly/Outlier Detection
• Machine Learning: Active Learning
• Reasoning under Uncertainty: Relational Probabilistic Models

Application Category

• Graph Algorithms and Modeling for the Web: Representation, reconstruction, and subgraph or motif discovery in Web-related graphs
• User Modeling, Personalization and Recommendation: Fairness-aware retrieval and ranking
• Web Mining and Content Analysis: Models for Web evolution

Abstract

Real-world observational data often contain existing or emerging heterogeneous subpopulations that deviate from global patterns. The majority of models tend to overlook these underrepresented groups, leading to inaccurate or even harmful predictions. Existing solutions often rely on detecting these samples as Out-of-domain (OOD) rather than adapting the model to new emerging patterns. We introduce DynaSubVAE, a Dynamic Subgrouping Variational Autoencoder framework that jointly performs representation learning and adaptive OOD detection. Unlike conventional approaches, DynaSubVAE evolves with the data by dynamically updating its latent structure to capture new trends. It leverages a novel non-parametric clustering mechanism, inspired by Gaussian Mixture Models, to discover and model latent subgroups based on embedding similarity. Extensive experiments show that DynaSubVAE achieves competitive performance in both near-OOD and far-OOD detection, and excels in class-OOD scenarios where an entire class is missing during training. We further illustrate that our dynamic subgrouping mechanism outperforms standalone clustering methods such as GMM and KMeans++ in terms of both OOD accuracy and regret precision.

Problem

• Detects heterogeneous subpopulations in observational data
• Adapts models to new emerging patterns dynamically
• Improves OOD detection accuracy for underrepresented groups

Innovation

• Dynamic Subgrouping VAE for adaptive OOD detection
• Non-parametric clustering based on embedding similarity
• Joint representation learning and latent structure updating