Convergence Acceleration for Multiobjective Sparse Reconstruction via Knowledge Transfer
Multiobjective sparse reconstruction (MOSR) methods can potentially obtain superior reconstruction performance. However, they suffer from high computational cost, especially in high-dimensional reconstruction. Furthermore, they are generally implemented independently without reusing prior knowledge from past experiences, leading to unnecessary computational consumption due to the re-exploration of similar search spaces. To address these problems, we propose a sparse-constraint knowledge transfer operator to accelerate the convergence of MOSR solvers by reusing the knowledge from past problem-solving experiences. Firstly, we introduce the deep nonlinear feature coding method to extract the feature mapping between the search of the current problem and a previously solved MOSR problem. Through this mapping, we learn a set of knowledge-induced solutions which contain the search experience of the past problem. Thereafter, we develop and apply a sparse-constraint strategy to refine these learned solutions to guarantee their sparse characteristics. Finally, we inject the refined solutions into the iteration of the current problem to facilitate the convergence. To validate the efficiency of the proposed operator, comprehensive studies on extensive simulated signal reconstruction are conducted.
KeywordsSparse reconstruction Multiobjective evolutionary algorithm Learning Knowledge transfer
This work was supported by the China Scholarship Council under Grant 201706540025.
- 3.Chen, M., Xu, Z., Weinberger, K., Sha, F.: Marginalized denoising autoencoders for domain adaptation. In: Proceedings of the 29th International Conference on Machine Learning (2012)Google Scholar
- 9.Gretton, A., Borgwardt, K.M., Rasch, M., Schölkopf, B., Smola, A.J.: A kernel method for the two-sample-problem. In: Proceedings of the Conference on Neural Information Processing System, pp. 513–520 (2007)Google Scholar
- 15.Liu, C., Zhao, Q., Yan, B., Elsayed, S., Ray, T., Sarker, R.: Adaptive sorting-based evolutionary algorithm for many-objective optimization. IEEE Trans. Evol. Comput. (in press). https://doi.org/10.1109/TEVC20182848254
- 19.Wei, P., Ke, Y., Goh, C.K.: Deep nonlinear feature coding for unsupervised domain adaptation. In: IJCAI, pp. 2189–2195 (2016)Google Scholar