A fast dynamic evolutionary multiobjective algorithm via manifold transfer learning
Contribution
Transfers solutions through learned manifolds, making high-dimensional dynamic optimization more tractable.
Abstract
Many dynamic optimization problems are high-dimensional, with hundreds or thousands of decision variables, and the underlying landscape shifts in low-dimensional manifolds rather than in the full ambient space. This paper proposes a fast dynamic evolutionary multiobjective algorithm that learns these manifolds from historical optimal populations and transfers solutions along them when a change is detected. By exploiting manifold structure, the algorithm achieves order-of-magnitude speedups over state-of-the-art baselines on high-dimensional DMOPs.
Why This Paper Matters
If you work on expensive or large-scale dynamic optimization — supply chains, dynamic pricing, large engineering systems — this paper shows how to scale transfer-based methods to settings where they previously did not work.
BibTeX
@article{jiang2020fast,
title={A fast dynamic evolutionary multiobjective algorithm via manifold transfer learning},
author={Jiang, Min and Wang, Zhenzhong and Qiu, Liming and Guo, Shihui and Gao, Xing and Tan, Kay Chen},
journal={IEEE Transactions on Cybernetics},
volume={51},
number={7},
pages={3417--3428},
year={2020},
publisher={IEEE},
}