Evaluating Efficient Performance Estimators of Neural Architectures

Authors: Xuefei Ning, Changcheng Tang, Wenshuo Li, Zixuan Zhou, Shuang Liang, Huazhong Yang, Yu Wang

NeurIPS 2021 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental In this paper, we conduct an extensive and organized assessment of OSEs and ZSEs on five NAS benchmarks: NAS-Bench-101/201/301, and NDS Res Net/Res Ne Xt-A. Specifically, we employ a set of NAS-oriented criteria to study the behavior of OSEs and ZSEs, and reveal their biases and variances.
Researcher Affiliation Collaboration Department of Electronic Engineering, Tsinghua University1 Novauto Technology Co. Ltd.2
Pseudocode No The paper does not contain any sections explicitly labeled as 'Pseudocode' or 'Algorithm', nor are there structured, code-like blocks describing a procedure.
Open Source Code Yes The code is available at https://github. com/walkerning/aw_nas [24].
Open Datasets Yes In this paper, we conduct an extensive and organized assessment of OSEs and ZSEs on five NAS benchmarks: NAS-Bench-101/201/301, and NDS Res Net/Res Ne Xt-A.
Dataset Splits Yes We inspect OSEs ranking quality when using different numbers of validation data batches to evaluate the OS scores, and find that on both NB201/NB301, using more data improves the estimation quality. Specifically, we compute the average OS accuracies over N validation batches, where each batch contains 128 examples.
Hardware Specification No The paper does not provide specific details about the hardware used for running the experiments, such as GPU models, CPU types, or memory.
Software Dependencies No The paper does not list specific software dependencies with version numbers (e.g., Python, PyTorch, TensorFlow versions).
Experiment Setup Yes Unless otherwise noted, MC sample S=1 is used in the experiments. And all training and evaluation settings are summarized in Appendix D. ... Specifically, we compute the average OS accuracies over N validation batches, where each batch contains 128 examples.