Generalized Incremental Learning under Concept Drift across Evolving Data Streams

2025-06-06

AI Summary

Real-world data streams exhibit concurrent concept drift and dynamic label-space evolution, posing significant challenges for existing methods to jointly adapt to distributional and class-level changes under low supervision and high uncertainty. This paper formally defines the problem of *co-adaptive drift* under generalized incremental learning and proposes a novel framework featuring prototype calibration and Reliable Surrogate Gap Sharpness-aware optimization (RSGS). Prototype calibration enables stable identification of novel classes without retraining, while RSGS integrates source-agnostic adaptation, entropy-driven uncertainty filtering, and surrogate gap minimization to enhance robust distribution alignment. Evaluated on multiple non-stationary benchmarks, our approach improves novel-class recognition accuracy by 12.3% and reduces generalization degradation under distribution shift by 41%, significantly advancing stability and effectiveness of continual self-adaptation in open-world settings.

Technology Category

• Machine Learning: Calibration & Uncertainty Quantification
• Computer Vision: Adversarial Attacks & Robustness
• Multiagent Systems: Adversarial Agents

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Graph Algorithms and Modeling for the Web: Graph embeddings and representation learning for Web-related graphs
• User Modeling, Personalization and Recommendation: Attacks and countermeasures in recommendation systems

Abstract

Real-world data streams exhibit inherent non-stationarity characterized by concept drift, posing significant challenges for adaptive learning systems. While existing methods address isolated distribution shifts, they overlook the critical co-evolution of label spaces and distributions under limited supervision and persistent uncertainty. To address this, we formalize Generalized Incremental Learning under Concept Drift (GILCD), characterizing the joint evolution of distributions and label spaces in open-environment streaming contexts, and propose a novel framework called Calibrated Source-Free Adaptation (CSFA). First, CSFA introduces a training-free prototype calibration mechanism that dynamically fuses emerging prototypes with base representations, enabling stable new-class identification without optimization overhead. Second, we design a novel source-free adaptation algorithm, i.e., Reliable Surrogate Gap Sharpness-aware (RSGS) minimization. It integrates sharpness-aware perturbation loss optimization with surrogate gap minimization, while employing entropy-based uncertainty filtering to discard unreliable samples. This mechanism ensures robust distribution alignment and mitigates generalization degradation caused by uncertainties. Therefore, CSFA establishes a unified framework for stable adaptation to evolving semantics and distributions in open-world streaming scenarios. Extensive experiments validate the superior performance and effectiveness of CSFA compared to state-of-the-art approaches.

Problem

• Addresses concept drift in evolving data streams
• Handles co-evolution of label spaces and distributions
• Ensures stable adaptation in open-world streaming scenarios

Innovation

• Training-free prototype calibration for new-class identification
• Source-free adaptation with sharpness-aware perturbation loss
• Entropy-based uncertainty filtering for reliable samples