GeNAS: Neural Architecture Search with Better Generalization
Authors: Joonhyun Jeong, Joonsang Yu, Geondo Park, Dongyoon Han, YoungJoon Yoo
IJCAI 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We evaluate our proposed method on various search spaces, showing similar or even better performance compared to the state-of-the-art NAS methods. Notably, the resultant architecture found by flatness measure generalizes robustly to various shifts in data distribution (e.g. Image Net-V2,-A,-O), as well as various tasks such as object detection and semantic segmentation. |
| Researcher Affiliation | Collaboration | Joonhyun Jeong1,2 , Joonsang Yu1,3 , Geondo Park2 , Dongyoon Han3 and Young Joon Yoo1 1NAVER Cloud, Image Vision 2KAIST 3NAVER AI Lab {joonhyun.jeong, joonsang.yu}@navercorp.com, geondopark@kaist.ac.kr, {dongyoon.han, youngjoon.yoo}@navercorp.com |
| Pseudocode | No | The paper provides mathematical equations but no structured pseudocode or algorithm blocks. |
| Open Source Code | Yes | 1Code is available at https://github.com/clovaai/Ge NAS. |
| Open Datasets | Yes | We first evaluate our proposed Ge NAS framework on widely used benchmark dataset, Image Net with DARTS [Liu et al., 2018b] search space. Furthermore, we thoroughly conduct ablation studies with regard to the components of Ge NAS on NAS-Bench-201 [Dong and Yang, 2020] benchmark. (...) MS-COCO [Lin et al., 2014] and Cityscapes [Cordts et al., 2016] dataset. |
| Dataset Splits | Yes | Specifically, we search architectures with 8 normal cells (i.e., stride = 1) and 2 reduction cells (i.e., stride = 2) on CIFAR-10/100, and transfer these normal / reduction cell architectures onto Image Net by training from scratch and evaluating top-1 accuracy on Image Net validation set. |
| Hardware Specification | Yes | We measured the execution time spent for the Super Net training and the search process, using a single NVIDIA V100 GPU. |
| Software Dependencies | No | We adopt the default training strategy of Retina Net [Lin et al., 2017] from Detectron2 [Wu et al., 2019]. (...) trained with MMSegmentation [Contributors, 2020] framework. No specific version numbers are provided for these software dependencies. |
| Experiment Setup | Yes | Specifically, we adjust α in Eq (4), where α = 0 denotes searching with only flatness of local minima. Results on Table 6 demonstrate that as α value increases from zero to one, search performance is drastically enhanced, indicating the indispensability of searching with both flatness and depth of minima. (...) Specifically, we adjust γ in Eq (5), which balances the coefficient concerning the ratio of flatness to angle term. |