 |
International Society of Science and Applied Technologies |
| Using A2C for Hyperparameter Optimization of Deep Learning Models | ||||
| Author |
Shun-Chang Lin
|
|||
| Co-Author(s) |
Gi-Wei Lin; Fang-Chih Tien
|
|||
| Abstract |
Deep learning has made remarkable advancements in machine learning, demonstrating exceptional performance in defect detection, object recognition, and image processing. AI models are widely adopted in industry due to their adaptability, efficiency, and accuracy. Reinforcement Learning (RL), a critical branch of machine learning, empowers algorithms to make optimal decisions through continuous interaction with the environment. This paper presents A2C-HOD(Advantage Actor- Critic-based Hyperparameter Optimization for Deep Neural Networks), an innovative method that applies the A2C reinforcement learning framework to hyperparameter optimization. We integrated A2C-HOD into the training pipeline of the YOLO model and evaluated its effectiveness on both standard and custom-built datasets. Experimental results demonstrate that A2C-HOD significantly enhances model performance and stability by efficiently identifying optimal hyperparameter configurations.
|
|||
| Keywords | Reinforcement Learning, Deep Learning, Hyperparameter optimization | |||
| Article #: RQD2025-115 | ||||
Proceedings of 30th ISSAT International Conference on Reliability & Quality in Design |