Efficient Neural Architecture Search via Proximal Iterations

Authors: Quanming Yao, Ju Xu, Wei-Wei Tu, Zhanxing Zhu6664-6671

AAAI 2020 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Here, we perform experiments with searching CNN and RNN structures. Four datasets, i.e., CIFAR-10, Image Net, PTB, WT2 will be utilized in our experiments (see Appendix B.1).
Researcher Affiliation Collaboration Quanming Yao,1 Ju Xu,3 Wei-Wei Tu,1 Zhanxing Zhu2,3,4 14Paradigm Inc, 2School of Mathematical Sciences, Peking University 3Center for Data Science, Peking University, 4Beijing Institute of Big Data Research (BIBDR) {yaoquanming, tuweiwei}@4paradigm.com, {xuju, zhanxing.zhu}@pku.edu.cn
Pseudocode Yes Algorithm 2 NASP: Efficient Neural Architecture Search with Proximal Iterations.
Open Source Code Yes The implementation of NASP is available at: https://github. com/xujinfan/NASP-codes.
Open Datasets Yes Four datasets, i.e., CIFAR-10, Image Net, PTB, WT2 will be utilized in our experiments (see Appendix B.1). ... we search architectures on CIFAR-10 ((Krizhevsky 2009)).
Dataset Splits Yes min A Lval (w , A) , s.t. w = arg min w Ltrain (w, A) ... where Lval (resp. Ltrain) is the loss on validation (resp. training) set...
Hardware Specification No The paper mentions "hundreds of GPU" and "GPU days" but does not specify exact GPU models or other hardware components used for their experiments.
Software Dependencies No The paper does not specify software dependencies with version numbers.
Experiment Setup Yes Following (Liu, Simonyan, and Yang 2019), the convolutional cell consists of N = 7 nodes, and the network is obtained by stacking cells for 8 times; in the search process, we train a small network stacked by 8 cells with 50 epochs (see Appendix B.2).