Publication: High-speed interconnect modelling by stacking ensemble learning method
Loading...
Date
2024-08
Authors
Chiang, Sze Kern
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
This thesis proposes the challenges and limitation of machine learning modelling techniques, for the problem inferring S-parameter of Via-Array in high
speed signal interconnection and applying large database in the process. These machine learning modelling techniques learns the data generation process to
efficiently construct more precise trained network that will be used to explored design space of package interconnect and its electrical performance. The focus of the project is to determine the most accurate kind of models by combining machine learning models include Artificial Neural Networks (ANN), Ensembles of Trees, Kernel Approximation Regression. The combination method that will be used is ensemble learning method, specifically stacking method which combines the predictions of numerous base models to get a final prediction with better performance. The project will also pursue different kind of techniques to solve some problem like mitigating overfitting in machine learning models and identify the time taken to train complex models. The accuracy and the performace of the proposed modelling technique are then investigated by comparing their results with another model previous did. From this project, the result that has the most accurate among all the machine learning are the meta-model of EOT&ANN, with the performance in mean RMSE of 0.0062 and accuracy of 72.675%. This models also had been tested to have a healthy training process with no significance evidence of overfitting.