Smart Pruning of Deep Neural Networks Using Curve Fitting and Evolution of Weights
Published in International Conference on Machine Learning, Optimization, and Data Science, 2022
This conference paper addresses the critical challenge of compressing deep neural networks to enhance their suitability for embedded and edge devices, where space and computational resources are limited. The proposed methods focus on smart pruning techniques, specifically utilizing curve fitting and the evolution of weights, to reduce the size and computational demands of neural networks without sacrificing accuracy.
The research questions conventional pruning methods and introduces two novel strategies—evolution of weights and smart pruning—demonstrating their effectiveness in making deep neural networks more efficient and robust against noise and overfitting. These techniques were tested on benchmark datasets, showing significant improvements over standard pruning mechanisms. The code for these methods has been made available online for public use, encouraging further exploration and application in real-world scenarios.
Recommended citation: Islam, A., & Belhaouari, S. B. (2022). "Smart Pruning of Deep Neural Networks Using Curve Fitting and Evolution of Weights." In International Conference on Machine Learning, Optimization, and Data Science (pp. 62--76). Springer.
Download Paper