Test-Time Discovery via Hashing Memory

2025-03-12

AI Summary

This work addresses open-world recognition under dynamic class evolution during testing, proposing a fully test-time adaptive framework for joint recognition of known and novel classes—without any model retraining. To simultaneously tackle real-time discovery of unknown classes and sustained discrimination of known ones, we introduce a novel hash-based memory mechanism: (i) a locality-sensitive hashing (LSH) scheme encoding feature magnitude and direction; (ii) a collaborative architecture integrating prototype- and LSH-based classifiers; and (iii) a hash-neighborhood-driven self-correction module to enhance discriminative robustness. Our method operates solely via online feature retrieval and dynamic clustering, requiring no parameter updates. Evaluated on multiple benchmarks, it significantly improves novel-class detection accuracy while preserving high precision on known classes. The approach is computationally efficient, stable under continuous distribution shifts, and inherently scalable to evolving class vocabularies.

Technology Category

• Machine Learning: Adversarial Learning & Robustness
• Computer Vision: Adversarial Attacks & Robustness
• Search and Optimization: Learning to Search

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Web Mining and Content Analysis: Models for Web evolution
• Graph Algorithms and Modeling for the Web: Representation, reconstruction, and subgraph or motif discovery in Web-related graphs

Abstract

We introduce Test-Time Discovery (TTD) as a novel task that addresses class shifts during testing, requiring models to simultaneously identify emerging categories while preserving previously learned ones. A key challenge in TTD is distinguishing newly discovered classes from those already identified. To address this, we propose a training-free, hash-based memory mechanism that enhances class discovery through fine-grained comparisons with past test samples. Leveraging the characteristics of unknown classes, our approach introduces hash representation based on feature scale and directions, utilizing Locality-Sensitive Hashing (LSH) for efficient grouping of similar samples. This enables test samples to be easily and quickly compared with relevant past instances. Furthermore, we design a collaborative classification strategy, combining a prototype classifier for known classes with an LSH-based classifier for novel ones. To enhance reliability, we incorporate a self-correction mechanism that refines memory labels through hash-based neighbor retrieval, ensuring more stable and accurate class assignments. Experimental results demonstrate that our method achieves good discovery of novel categories while maintaining performance on known classes, establishing a new paradigm in model testing. Our code is available at https://github.com/fanlyu/ttd.

Problem

• Addresses class shifts during testing
• Identifies emerging categories while preserving learned ones
• Enhances class discovery via hash-based memory mechanism

Innovation

• Hash-based memory mechanism for class discovery
• Locality-Sensitive Hashing for efficient sample grouping
• Collaborative classification strategy combining prototype and LSH classifiers