The training phase of hotspot detection based on Machine Learning is time-consuming. In this paper
a parallel PCA algorithm based on MPI and a parallel AdaBoost algorithm based on OpenMP are proposed to reduce the training time. Firstly
the QR decomposition is used to optimize SVD
and then the MPI is used to implement the PCA parallel algorithm. Finally
the reduced dimension data is trained using multi-core CPU parallel algorithm. The experimental results show that the PCA parallel algorithm could achieve a 4.7 times speed-up ratio
and the AdaBoost parallel algorithm could achieve a 4.9 times speed-up ratio
which verifies the feasibility of parallelization.
关键词
Keywords
references
Quinn M J. MPI与OpenMP并行程序设计:C语言版[M].陈文光, 武永卫, 译.北京:清华大学出版社, 2004.
Drmac Z. A posteriori computation of the singular vectors in a preconditioned Jacobi SVD algorithm[J]. Ima Journal of Numerical Analysis, 1999, 19(2):191-213. DOI: 191
Matsunawa T, Gao J R, Yu B, et al. A new lithography hotspot detection framework based on AdaBoost classifier and simplified feature extraction[J]. Spie Advanced Lithography, 2015 :94270S-94270S-11. DOI: 94270S
Yu Y T. Machine-learning-based hotspot detection usingtopological classification and critical feature extraction[J]. IEEE/ACM Design Automation Conference(DAC), 2013, 34 (3) :1-6. DOI: 1
Wuu J, Pikus Y. Detecting context sensitive hotspots in standard cell libraries[J]. Proc. of SPIE, 2009, (7275):265-271. DOI: 265
Yoav F Schapire, Robert E. A decision-theoretic generalization of on-line learning and an application to boosting[J]. Journal of Computer and System Sciences, 1997, 55(1): 119-139. DOI: 119