GeM-EA: A Generative and Meta-learning Enhanced Evolutionary Algorithm for Streaming Data-Driven Optimization
Summary: arXiv:2604.12336v1 Announce Type: cross
Abstract
Streaming Data-Driven Optimization (SDDO) problems arise in many applications where data arrive continuously, and the optimization environment evolves over time. Concept drift produces non-stationary landscapes, making optimization methods challenging due to outdated models. Existing approaches often rely on simple surrogate combinations or directly injecting solutions, which may cause negative transfer under sudden environmental changes. We propose GeM-EA, a Generative and Meta-learning Enhanced Evolutionary Algorithm for SDDO that unifies meta-learned surrogate adaptation with generative replay for effective evolutionary search. Upon detecting concept drift, a bi-level meta-learning strategy rapidly initializes the surrogate using environment-relevant priors, while a linear residual component captures global trends. A multi-island evolutionary strategy further leverages historical knowledge via generative replay to accelerate optimization. Experimental results on benchmark SDDO problems demonstrate that GeM-EA achieves faster adaptation and improved robustness compared with state-of-the-art methods.
Introduction
In an era where data streams continuously, optimizing these streams presents unique challenges, particularly due to concept drift. This phenomenon occurs when the statistical properties of the target variable change, which can lead to outdated models that fail to adapt. Traditional optimization methods struggle to maintain efficacy in such dynamic environments.
Challenges in Streaming Data-Driven Optimization
- Non-stationary landscapes: The optimization environment is constantly evolving, causing traditional algorithms to perform suboptimally.
- Outdated models: As data continues to stream in, existing models may become irrelevant, leading to poor decision-making.
- Negative transfer: Methods that depend on simple surrogate combinations or direct solution injections can adversely affect performance during sudden changes.
The GeM-EA Solution
GeM-EA addresses the aforementioned challenges by integrating two powerful strategies: generative replay and meta-learning. This innovative approach allows for effective evolutionary search, ensuring that the algorithm can adapt rapidly to changing environments.
Key Components of GeM-EA
- Bi-level Meta-learning: This strategy quickly initializes the surrogate model utilizing environment-relevant priors, enabling rapid adaptation when concept drift is detected.
- Linear Residual Component: This component captures global trends, helping the algorithm remain aware of overarching patterns despite local changes.
- Multi-island Evolutionary Strategy: By leveraging historical knowledge through generative replay, this strategy enhances the optimization process, allowing for faster convergence.
Experimental Results
Through rigorous testing on benchmark SDDO problems, GeM-EA has demonstrated significant improvements in both adaptation speed and robustness. The results indicate that this novel algorithm outperforms state-of-the-art methods, providing a reliable solution for dynamic optimization challenges.
Conclusion
GeM-EA represents a significant advancement in the field of Streaming Data-Driven Optimization. By combining generative and meta-learning techniques, it offers a robust framework for adapting to continuous data streams and evolving environments. Future research may explore further enhancements and applications of this algorithm across various domains.