Publication: Fast transient simulations using temporal convolutional networks with an adaptive successive halving algorithm for automated hyperparameter optimizations
Loading...
Date
2023-10-01
Authors
Goay Chan Hong
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
This research presents a fast transient simulation method for high-speed channels using temporal neural networks (TCNs), which outperform recurrent architectures in terms of training speed and convergence. To optimize the TCN's performance, the adaptive successive halving algorithm (ASH-HPO) and its alternate version, ASH-HPO Ver2, are proposed for automated hyperparameter optimization (HPO). Results indicate that the TCN model trains 1.42 to 3.55 times faster than other converging models and achieves convergence across all tested high-speed channels. Furthermore, the ASH-HPO algorithm exhibits faster convergence compared to state-of-the-art methods such as Bayesian optimization (BO), successive halving, and hyperband (with speedup factors ranging from 1.04 to 7.90). Notably, the ASH-HPO Ver2 algorithm achieves even greater speed improvements, with training times approximately 1.07 to 2.40 times faster than the original version. These findings highlight the effectiveness of TCNs and the proposed ASH-HPO algorithms in accelerating high-speed channel simulations, offering promising advancements in transient analysis, enabling efficient and accurate simulations for high-speed channels, and facilitating improved understanding and design of such systems