SiameseDuo++: Active learning from data streams with dual augmented siamese networks

2025-03-01

Neurocomputing

AI Summary

Addressing the core challenges of pronounced concept drift, severe label scarcity, and strict memory constraints in data stream mining, this paper proposes a dual-path enhanced twin active learning framework tailored for dynamic streams. Methodologically, it introduces a novel dual-augmentation twin network architecture that jointly models sample similarity and predictive uncertainty, enabling unsupervised representation alignment and active query optimization in an integrated manner. Additionally, a streaming-adaptive thresholding mechanism is incorporated to support online concept drift detection and efficient, real-time sample selection. Evaluated on six real-world data stream benchmarks, the framework reduces labeling budget by 32% on average while improving F1-score by 11.7%, significantly outperforming existing state-of-the-art streaming active learning approaches.

Technology Category

• Machine Learning: Active Learning
• Data Mining & Knowledge Management: Data Stream Mining
• Search and Optimization: Sampling/Simulation-based Search

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Web Mining and Content Analysis: Content-based information diffusion
• Graph Algorithms and Modeling for the Web: Efficient manipulation of static and dynamic Web-related graphs

Problem

• Active learning for data streams with limited labeling budget
• Handling concept drift in nonstationary streaming data
• Real-time processing with constrained memory resources

Innovation

• Active learning with dual Siamese networks
• Latent space augmentation and strategy
• Memory-efficient real-time data processing