Earticle

현재 위치 Home

Convergence of Internet, Broadcasting and Communication

Beta and Alpha Regularizers of Mish Activation Functions for Machine Learning Applications in Deep Neural Networks

첫 페이지 보기
  • 발행기관
    국제인공지능학회(구 한국인터넷방송통신학회) 바로가기
  • 간행물
    International Journal of Internet, Broadcasting and Communication 바로가기
  • 통권
    Vol.14 No.1 (2022.02)바로가기
  • 페이지
    pp.136-141
  • 저자
    Peter Beatus Mathayo, Dae-Ki Kang
  • 언어
    영어(ENG)
  • URL
    https://www.earticle.net/Article/A409185

※ 원문제공기관과의 협약기간이 종료되어 열람이 제한될 수 있습니다.

원문정보

초록

영어
A very complex task in deep learning such as image classification must be solved with the help of neural networks and activation functions. The backpropagation algorithm advances backward from the output layer towards the input layer, the gradients often get smaller and smaller and approach zero which eventually leaves the weights of the initial or lower layers nearly unchanged, as a result, the gradient descent never converges to the optimum. We propose a two-factor non-saturating activation functions known as Bea-Mish for machine learning applications in deep neural networks. Our method uses two factors, beta (β) and alpha (α), to normalize the area below the boundary in the Mish activation function and we regard these elements as Bea. Bea-Mish provide a clear understanding of the behaviors and conditions governing this regularization term can lead to a more principled approach for constructing better performing activation functions. We evaluate Bea-Mish results against Mish and Swish activation functions in various models and data sets. Empirical results show that our approach (Bea-Mish) outperforms native Mish using SqueezeNet backbone with an average precision (AP50val) of 2.51% in CIFAR-10 and top-1accuracy in ResNet-50 on ImageNet-1k. shows an improvement of 1.20%.

목차

Abstract
1. Introduction
2. Theory
3. Experiment Settings
4. Results and Discussion
4.1 CIFAR-10 Dataset on Various Baseline Activation Functions
4.2 CIFAR-10 Dataset on Various Standard Neural Network Architectures
4.3 ImageNet-1k Dataset on Various Standard Neural Network Architectures
4.4 Ablation Study of α and β on CIFAR-10
5. Conclusion
Acknowledgement
References

저자

  • Peter Beatus Mathayo [ Master Student, Department of Computer Engineering, Graduate School, Dongseo University, Korea ]
  • Dae-Ki Kang [ Professor, Department of Computer Engineering, Dongseo University, Korea ] Corresponding author

참고문헌

자료제공 : 네이버학술정보

간행물 정보

발행기관

  • 발행기관명
    국제인공지능학회(구 한국인터넷방송통신학회) [The International Association for Artificial Intelligence]
  • 설립연도
    2000
  • 분야
    공학>전자/정보통신공학
  • 소개
    인터넷방송, 인터넷 TV , 방송 통신 네트워크 및 관련 분야에 대한 국내는 물론 국제적인 학술, 기술의 진흥발전에 공헌하고 지식 정보화 사회에 기여하고자 한다.

간행물

  • 간행물명
    International Journal of Internet, Broadcasting and Communication
  • 간기
    계간
  • pISSN
    2288-4920
  • eISSN
    2288-4939
  • 수록기간
    2009~2025
  • 십진분류
    KDC 326 DDC 380

이 권호 내 다른 논문 / International Journal of Internet, Broadcasting and Communication Vol.14 No.1

    피인용수 : 0(자료제공 : 네이버학술정보)

    함께 이용한 논문 이 논문을 다운로드한 분들이 이용한 다른 논문입니다.

      페이지 저장