NAFBee: Neural Network Activation Function Benchmark

This is a benchmark for networks with a variety of activation functions. NAFBee provides network information and accuracy. User can obtain the accuracy without training. NAFBee is used for RBFleX-NAS.

🟨 Download the paper

RBFleX-NAS: Training-Free Neural Architecture Search Using Radial Basis Function Kernel and Hyperparameter Detection
Tomomasa Yamasaki; Zhehui Wang; Tao Luo; Niangjun Chen; Bo Wang
IEEE Transactions on Neural Networks and Learning Systems (TNNLS)

🟨 Requirement

• python 3.x
• PyTorch

conda create -n myenv python=3.8
conda activate myenv
conda install -c conda-forge transformers=4.5.0 tokenizers=0.10.3
conda install -c conda-forge huggingface_hub
conda install pandas
conda install scikit-learn

🟨 Datasets for this neural network benchmark

• CIFAR-10 for VGG19
• SST-2 for BERT

🟨 How to use

0. Import packages

# VGG19
import json
from models import *

# BERT
import json
from BERT_model import BertModel

1. Load NAFBee.json

file_path = "NAFBee_VGG19.json" #or "NAFBee_BERT.json"
with open(file_path, "r") as json_file:
    nafbee = json.load(json_file)

2. Get the network information

info = nafbee["1"] #you can input numbers from 1 to 11.
print(info)

#{'network': 'VGG19', 'activation': 'ReLU', 'accuracy': 91.06}

3. All information of the No.1 network

info_network = info["network"]
info_activation = info["activation"]
info_accuracy = info["accuracy"]

4. Define the model on PyTorch

if "VGG" in info_network:
    model = VGG(info_network, info_activation)

🟨 Demo

You can see a program to get the model. You can add any program using the model on Pytorch such as training or scoring.

python NAFBee_VGG19.py #VGG19

python NAFBee_BERT.py #BERT

🟨 Citing RBFleX-NAS

If you use NAFBee, please cite the following paper:

@ARTICLE{10959729,
  author={Yamasaki, Tomomasa and Wang, Zhehui and Luo, Tao and Chen, Niangjun and Wang, Bo},
  journal={IEEE Transactions on Neural Networks and Learning Systems}, 
  title={RBFleX-NAS: Training-Free Neural Architecture Search Using Radial Basis Function Kernel and Hyperparameter Detection}, 
  year={2025},
  volume={36},
  number={6},
  pages={10057-10071},
  keywords={Accuracy;Training;Vectors;Kernel;Feature extraction;Correlation;Computer architecture;Benchmark testing;Predictive models;Neural architecture search;Activation;neural architecture search (NAS);radial basis function (RBF);training-free},
  doi={10.1109/TNNLS.2025.3552693}}

🟨 Licence

MIT Licence