SphOR: A Representation Learning Perspective on Open-set Recognition for Identifying Unknown Classes in Deep Learning Models

2025-03-11

AI Summary

To address the challenges of unreliable unknown-class rejection and high computational/training overhead in open-set recognition (OSR), this paper proposes a lightweight representation learning framework based on spherical embedding. Methodologically, it employs end-to-end trainable spherical feature representations modeled probabilistically via a von Mises–Fisher (vMF) mixture distribution—a novel integration of spherical geometry and directional statistics. It further introduces semantically ambiguous sample augmentation to sharpen decision boundaries for unknown classes, a first in OSR literature. The work also establishes an intrinsic connection between spherical geometric structure and OSR performance, providing theoretical insight into representation design. Evaluated on multiple standard OSR benchmarks, the method achieves state-of-the-art results, improving unknown-class detection accuracy by up to 6% while significantly reducing both inference latency and training cost compared to existing approaches.

Technology Category

• Machine Learning: Calibration & Uncertainty Quantification
• Computer Vision: Representation Learning for Vision
• Reasoning under Uncertainty: Uncertainty Representations

Application Category

• Graph Algorithms and Modeling for the Web: Graph embeddings and representation learning for Web-related graphs
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Semantics and Knowledge: Representation, semantic annotation, enhancement, enrichments, access and/or integration of a variety of data on the Web

Abstract

The widespread use of deep learning classifiers necessitates Open-set recognition (OSR), which enables the identification of input data not only from classes known during training but also from unknown classes that might be present in test data. Many existing OSR methods are computationally expensive due to the reliance on complex generative models or suffer from high training costs. We investigate OSR from a representation-learning perspective, specifically through spherical embeddings. We introduce SphOR, a computationally efficient representation learning method that models the feature space as a mixture of von Mises-Fisher distributions. This approach enables the use of semantically ambiguous samples during training, to improve the detection of samples from unknown classes. We further explore the relationship between OSR performance and key representation learning properties which influence how well features are structured in high-dimensional space. Extensive experiments on multiple OSR benchmarks demonstrate the effectiveness of our method, producing state-of-the-art results, with improvements up-to 6% that validate its performance.

Problem

• Open-set recognition for identifying unknown classes
• Computationally efficient representation learning method
• Improving detection of unknown class samples

Innovation

• Uses spherical embeddings for representation learning
• Models feature space with von Mises-Fisher distributions
• Improves unknown class detection with ambiguous samples