Bridging the Gap between Sample-based and One-shot Neural Architecture Search with BONAS

Authors: Han Shi, Renjie Pi, Hang Xu, Zhenguo Li, James Kwok, Tong Zhang

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

Reproducibility Variable Result LLM Response
Research Type Experimental Extensive experiments are conducted to verify the effectiveness of our method over competing algorithms.1
Researcher Affiliation Collaboration Han Shi1 , Renjie Pi2 , Hang Xu3, Zhenguo Li3, James T. Kwok1, Tong Zhang1 1Hong Kong University of Science and Technology, Hong Kong {hshiac,jamesk}@cse.ust.hk, tongzhang@ust.hk 2The University of Hong Kong, Hong Kong pipilu@hku.hk 3Huawei Noah s Ark Lab {xu.hang,li.zhenguo}@huawei.com
Pseudocode Yes Algorithm 1 Generic BO procedure for NAS. and Algorithm 2 BONAS.
Open Source Code Yes 1The code is available at https://github.com/pipilurj/BONAS.
Open Datasets Yes In the following experiments, we use NAS-Bench-101 [38], which is the largest NAS benchmark data set (with 423K convolutional architectures), and the more recent NAS-Bench-201 [8], which uses a different search space (with 15K architectures) and is applicable to almost any NAS algorithm. and a new NAS benchmark data set LSTM-12K we recently collected for LSTMs. and on the Penn Tree Bank data set [21] and CIFAR-10 data set. and Image Net [6].
Dataset Splits Yes For each data set, we use 85% of the data for training, 10% for validation, and the rest for testing.
Hardware Specification Yes All experiments are performed on NVIDIA Tesla V100 GPUs.
Software Dependencies No The paper mentions software like 'Adam optimizer' but does not provide specific version numbers for any key software components or libraries.
Experiment Setup Yes The GCN has four hidden layers with 64 units each. Training is performed by minimizing the square loss, using the Adam optimizer [12] with a learning rate of 0.001 and a mini-batch size of 128. and In step 10 of Algorithm 2, k = 100 models are merged to a super-network and trained for 100 epochs using the procedure discussed in Section 3.2.