| | | 20,000+ Fresh Resumes Monthly | |
|
|
| |
Click here or scroll down to respond to this candidate Candidate's Name
Tel: (+1)Street Address -719-6942 | Email: EMAIL AVAILABLE | LinkedIn: https://LINKEDIN LINK AVAILABLE
I m applying for a position of Software Engineer. I believe it s my enthusiasm about coding, critical thinking, and my curiosity in
understanding how things work and why they work that way make me distinguished. Knowing how to solve a problem and why a solution
works is more important than solving it. With my communication skills and passion for technology, I believe I am a reliable teammate.
EDUCATION
Columbia University New York, NY, US
Master of Science in Computer Science 05/2022
Sun Yat-sen University Guangzhou, Guangdong, China
Bachelor of Engineering in Intelligence Science and Technology 06/2020
Awards: First Prize Scholarship (top 10%), First Prize in ACM Academy Novice Competition (top 10%)
Projects: Development of Embedded Face Recognition System (Provided the basis for the access control system of school)
PROFESSIONAL EXPERIENCE
PAX Technology Jacksonville, FL, US
Software Engineer 01/2023-present
Developed Basil System for the warehouse project with Spring Boot and Vue.js framework. Implemented functions in service
layer to finish the operations of CRUD in databases for recording and managing the equipment returned by customers and the
materials needed for repairs.
Designed and implemented the email format for sending email alerts to users who register and change their information using the
APIs provided by Syspro and asynchronous programming.
Most of the current database query statements are string splicing statements, which lack a string filtering mechanism and have
potential SQL injection problems. To Solve the problem, I used MyBatis for parameterized queries and refactored codes of the
query part, improving the security.
Some redundant columns of data tables and large amounts of data add a lot of overhead to the query. I optimized the structure of
the data table to remove the redundant columns and optimized the query statements using sharding and multi-threading
techniques, which saved 30% of time.
Due to the increasing volume of data and the fact that there are far more queries than writes to the data, the load pressure on the
database also becomes higher. To improve both writing and query performance, I implemented the technology of Read/Write
Splitting, making application write requests to the database only on the master database and read requests on the slave database.
The implementation reduced query time by 14%, increased throughput by 11% and reduced database load.
Cadence Design Systems Shanghai, China
Software Engineer Intern 05/2021-08/2021
I had the opportunity to delve into semiconductor development, specifically focusing on optimizing a grid-based router structure
at the software level. In the development of EDA software Virtuoso, which assists in the definition, planning, design,
implementation, verification and subsequent fabrication of semiconductor devices or chips, one of the key tasks involved the
modification of M0 jumpers to enable the sources to avoid high resistance, ultimately improved the efficiency of circuit flows by
19%, which is beneficial to the software of PCB design.
Additionally, I was assigned the responsibility of implementing an algorithm to find the shortest route between two points in
large-scale nets. Upon careful consideration, I proposed the adoption of the A-star Algorithm to replace the Breadth-First Search
(BFS) algorithm. The decision was based on the suitability of the Euclidean distance for estimating the function, considering the
nature of semiconductor development. I actively applied the A-star Algorithm with pruning to enhance the efficiency of the routing
algorithm.
Recognizing that the A-star Algorithm on large-scale nets incurred a significant time cost, I took further action to optimize the
data structure. Specifically, I replaced the existing multimap with a priority queue to store the state of each net. This optimization
was driven by the realization that only the minimum estimated distance was crucial for the algorithm's success. As a result, this
enhancement led to a 6% reduction in processing time.
DiDi AI Labs Beijing, China
Algorithm Engineer Intern 07/2020-12/2020
Conducted research related to Auto Machine Learning, especially on Neural Architecture Search (NAS) and DARTS to help
the team calculate the feasibility of each model ported to mobile device.
Reproduced different models with Pytorch and conducted performance benchmarking based on top-1 and top-5 accuracy on
ImageNet or Cifar-10, in terms of parameter sizes, and training time of GPU hours.
Since the current indicators are too homogenous, it s unfair to evaluate a model based on its accuracy merely. I figured out new
metrics like search space of width and depth, and types of blocks to develop the optimal model for mobile computing that improved
11% overall performance. The model strikes a good balance between performance, number of parameters, and training time.
The laboratory aimed to deploy the model to mobile devices. I summarized characteristics of each model and wrote an overview,
providing the laboratory with a theoretical basis for research on automatic driving in the future.
PROJECT EXPERIENCE
Development of Embedded Face Recognition System 09/2018-12/2019
Our lab aimed to develop a system of recognition for all students on campus. A good neural network needs a lot of data to support
it, so I looked for a dataset with coordinates of key points of the human skeleton. By estimating the proportions of each part of the
human body, I designed algorithms to filter the data so that we only need to manually remove a small amount of data, which
greatly reduces the workload.
To avoid the time-consuming task of training a new model from scratch, we decided to use the pre-trained VGG-16 model as the
backbone, which I benchmarked on our own dataset, with a recognition accuracy of about 40 percent.
Due to the limited performance of embedded mobile, neural network models trained on GPUs in general cannot be run directly
on mobile, so we compressed the model using model distillation and model pruning techniques, and then trained it on our own
dataset, which resulted in more than 90% accuracy. Finally, we deployed the model and the parameters into the embedded camera.
To verify the performance of the model in practical applications, we took the lead in testing this system in the class. We first
stored the ID photos as well as the ids of the students in the class in the database as a test set, and then we performed face
recognition, comparing the faces obtained on site with all the human connections in the database, and selecting the ones with the
highest similarity and greater than 0.6 as the answers. After several fine-tuning, the recognition correctness rate was increased
from 82% to 100%. I also participated in the development of backend which called the model for calculations using framework
Flask in Python. The school's current access control system is based on our embedded face recognition system.
BuzzCar: Used Car Platform 09/2023-12/2023
Utilized UML Class Diagrams to design Figma of all objects.
Designed and implemented database with SQL, designed RESTful API rules for URLs.
Implemented backend development of all APIs with Java using Spring Boot.
Utilized JavaScript called APIs in frontend and presented data with HTML and CSS.
Wrote unit test codes and tested the output with Postman.
Segmentation of White Matter Hyperintensities 07/2019-09/2019
Conducted segmentation of brain FLAIR and T1 scan images using neural network models of DeepMedic, FCN and ResNet
with Keras, and marked the part of white matter hyperintensities (WMH).
Pre-processed the same raw data by removing masks in different parts like skulls and tested the effects that removal of skull masks
improved Recall by 5% for FCN and F1-score by 10% for DeepMedic.
Implemented the models of FCN and DeepMedic on images of patients with size 224*224 to identify white matter hyperintensities,
which largely saved preparation effort and increased diagnostic accuracy.
SKILLS
Programming Languages: C++, Python, Java, SQL, MATLAB, JavaScript, Scala, QT
Framework Tools: Spring Boot, MyBatis, Tensorflow, Pytorch, Keras, Vue.js, Flask
Development Tools: Visual Studio, Pycharm, Jupyter Notebook, Docker, Git, Anaconda, QT Creator, Postman, Datagrip, MySQL
Knowledges: Data Structures, Algorithms, Machine Learning, Calculus, Probability Theory
|
