Effect of the Slope of Symmetric Saturated Activation Functions on Deep Learning

10th International Scientific Conference Technics, Informatics and Education – TIE 2024, str. 79-82

АУТОР(И) / AUTHOR(S): Maja Lutovac Banduka , Vladimir Poučki , Vladimir Mladenović , Miroslav Lutovac 

Download Full Pdf  

DOI: 10.46793/TIE24.079LB

САЖЕТАК /ABSTRACT:

It is presented how the slope of symmetric activation functions with saturation affects class detection using symbolic analysis. Different activation functions can be used to increase the most likely detected classes. The main result is the determination of the highest slope of the activation function and the lowest slope of the activation function in terms of the number of neurons in the layer

КЉУЧНЕ РЕЧИ / KEYWORDS: 

class detection; probability; automated drawing; symbolic solving of the neural network

ЛИТЕРАТУРА / REFERENCES:

  1. Lutovac Banduka, M., Franc, I., Milićević, V., Zdravković, N., & Dimitrijević N. (2023). Symbolic analysis of classical neural networks for deep learning, 2023110446, Online doi.org/10.20944/preprints202311.0446.v1
  2. Bernard, E. (2022). Introduction to machine learning, Champaign, IL, USA: Wolfram Media.
  3. Lutovac Banduka, M., Milosevic, D., Cen, Y., Kar, A., & Mladenovic, V. (2023). Graphical user interface for design, analysis, validation, and reporting of continuous-time systems using Wolfram language, JCSC, #2350244, 32(14), 1–15.
  4. Wolfram, S. (2023). An elementary introduction to the Wolfram language, 3rd Champaign, IL: Wolfram Media.
  5. Lutovac Banduka, M., Simović, A., Orlić, V., & Stevanović, A. (2023). Dissipation minimi­zation of two-stage amplifier using deep learning, Serbian Journal of Electrical Engineering, 20(2), 129–145.
  6. Milićević, V., Lutovac Banduka, M., Franc, I., Zdravković, N., & Dimitrijević, N. (2025). Symbolic analysis of classical neural networks for deep learning, International Journal for Quality Research, 19(1), in press.
  7. Lutovac Banduka, M., Franc, I., Milićević, V., Zdravković, N., & Dimitrijević N. (2024). The effect of the number of hidden layers and activation functions on deep learning, unpublished.
  8. Lutovac Banduka, M., Lutovac, M. (2024). How to design multiplierless neural networks for deep learning?, EasyChair Preprint #13105, 1-4.
  9. Lutovac Banduka, M., Lutovac, M. (2024). Multiplierless neural networks for deep learning, Mediterranean Conference on Embedded Computing (MECO), 11-14 June 2024, Budva, 262-265.