Page History: Derivative-free optimization by classification
Compare Page Revisions
Page Revision: 2016/12/09 13:57
[
Back to main page ]
For optimization non-covnex and complex functions, derivative-based methods may not effective because a point-wise derivative does not reflect the global landscape of the function. Instead, sampling in the solution space can reveal some global information about the function, and thus sampling-based methods, such as evolutionary algorithms, can be more suitable for complex optimizations.
RACOS
RACOS is designed according to the general complexity upper bound of a sampling-and-learning framework. It can be used to optimize functions in bounded continuous, discrete, and mixed solutions space. For details please see:
Yang Yu, Hong Qian, and Yi-Qi Hu. Derivative-free optimization via classification. In: Proceedings of the 30th AAAI Conference on Artificial Intelligence (AAAI'16), Phoenix, AZ, 2016. (
PDF) (
Appendix)
Sequential RACOS
Sequential RACOS (SRACOS) is the online version of RACOS, which updates the model after every sample of solution. In online scenarios, where solutions have to be evaluated one after another, SRACOS can converge much faster than RACOS.
Yi-Qi Hu, Hong Qian, and Yang Yu. Sequential classification-based optimization for direct policy search. In: Proceedings of the 31st AAAI Conference on Artificial Intelligence (AAAI’17), San Francisco, CA, 2017. (
PDF with Appendix)
The latest version is in Github (available in Java, Matlab, C++, and Python):
https://github.com/eyounx/RACOS
Local download:
Java version (used in the experiments): (
Code Download in Zip, 32KB)
Matlab version: (
Code Download in Zip, 10KB)
- The codes are released under the GNU GPL 2.0 license. For commercial purposes, please contact me or Prof. Zhi-Hua Zhou.