Back to search

Georgia Institute of Technology

DEEP LEARNING IN BIOMEDICAL INFORMATICS AND MODERN CIRCUIT DESIGN

Abstract

dc:description.abstract

Deep learning has grown extensively in recent years. The application of deep learning in many fields has improved substantially the performance in computer vision, speech recognition and natural language processing. In my research, we leverage deep learning knowledge and framework for COVID-19 related key measurements prediction and circuit design. The global spread of COVID-19, the disease caused by the novel coronavirus SARS-CoV-2, has casted a significant threat to mankind. As the COVID-19 situation continues to evolve, predicting localized disease severity is crucial for advanced resource allocation. Authorities have the urgent need to make informed decision on the resource allocation and related policies, based on the COVID-19 related key measurements. In response to the need of accurate predictions on the key measurements, we develop a learning framework for the prediction of the COVID-19 related mortality and number of hospitalizations. Our model utilizes publicly available information for COVID-19 related confirmed cases, deaths, community mobility and demographic information to produce the county and state level number of deaths prediction. Furthermore, by incorporating of current week’s numbers of hospitalizations and the median income as the input data, the model can be used to predict the number of hospitalizations for the next four weeks. We design our model to incorporate a self attention model, known as the transformer model to capture both short term and long term dependencies within the temporal data. Our model benefits from the mixup data augmentation and residual connection to produce accurate prediction, when compare among benchmark models at the Centers for Disease Control and Prevention (CDC)’s website. In circuit design, we are solving a design problem that involves the direct end-to-end generation of RF/mm-wave passive matching network with 3D EM structures. Traditionally, such a circuit design is a designer in the loop process, with stringent demand on the designers to have strong electrical knowledge and year of experiences. Even so, the design process is computational intensive and iterative. By looking at the design process as an inverse problem, we propose a novel approach that leverage modern deep learning based framework to effectively and efficiently solve this problem. Our model incorporates a shared encoder that capture the intrinsic relationship among the circuit parameters, geometrical parameters and the performance parameters. This allows an accurate prediction of the geometrical parameters for the crucial transformer component used in the passive matching network. Furthermore, our learning framework is a highly flexible and general framework. A well trained model from our framework can be further used as a pretrained model, and finetuned for other similar networks. With such a impactful learning framework, this is a breakthrough in the circuit design field and has unlimited potential to be applied for a variety of circuit design problems.

Degree

thesis:*
Level thesis:degree_level
Doctoral
Department dc:contributor.department
Industrial and Systems Engineering
Grantor dc:publisher
Georgia Institute of Technology
Year dc:date.issued
2022

Author and committee

dc:creator, dc:contributor.*
Author dc:creator
  • Er, Siawpeng
Advisors dc:contributor.advisor
  • Yang, Shihao
  • Zhao, Tuo
Committee members dc:contributor.committeemember
  • Jordan, King
  • Wang, Hua
  • Zhang, Chao

Subjects

dc:subject × 4

Rights

Language dc:language.iso
en_US

Identifiers

dc:identifier.*
Handle dc:identifier.uri
http://hdl.handle.net/1853/67320

Chain of custody

source
Harvested from
Georgia Tech
Base URL
repository.gatech.edu/server/oai/request
Last updated
2026-11-19
Source record
OAI-PMH GetRecord
citation

Er, Siawpeng. DEEP LEARNING IN BIOMEDICAL INFORMATICS AND MODERN CIRCUIT DESIGN. Doctoral thesis, Georgia Institute of Technology, 2022. http://hdl.handle.net/1853/67320