We introduce a NAS benchmark for text classification task, consisting of 4k ECGP-evolved architectures on MR, 3.4k on SST2, and 3.5k on SST5, along with their respective performances. By leveraging a designated function set, including Attention, Convolution, GRU, etc., the top-performing architecture achieves accuracy rates of 80%, 85%, and 47% on MR, SST2, and SST5, respectively.
If you find our Benchmark useful, please cite our paper "Neural Architecture Search for Text Classification With Limited Computing Resources Using Efficient Cartesian Genetic Programming":
@article{wu_neural_2023,
title = {Neural Architecture Search for Text Classification with Limited Computing Resources Using Efficient Cartesian Genetic Programming},
journal = {{IEEE} {Transactions} on {Evolutionary} {Computation} (in press)},
author = {Wu, Xuan and Wang, Di and Chen, Huanhuan and Yan, Lele and Xiao, Yubin and Miao, Chunyan and Ge, Hongwei and Xu, Dong and Liang, Yanchun and Wang, Kangping and Wu, Chunguo and Zhou, You},
}