CV

Awards and Accomplishments at UofT:
<sub>- Overall GPA: 3.89/4, overall course average: 90.3/100, computer engineering ranking top 7 over 179.
- Dean’s Honor List all semesters.
- John M. Empey Scholarship in 2020.
- UTEA (University of Toronto Excellence Award) Scholarship in 2021.
- Certificate of Distinction for Excellence in Design Project in 2022.</sub>

Technical Skills:
<sub>- Back-end Software Development: C / C++ / Java / Android Development / SQL
- Machine Learning: Python (Pytorch / Tensorflow / Pandas / OpenCV / Keras / Objax)
- OS and Version Control: Development based on Unix and Linux / Git
- Others: Matlab / Verilog / Assembly / LaTeX / Microsoft Office</sub>

Publications:
_{[1] “EDDA: Explanation-driven Data Augmentation to improve Model and Explanation Faithfulness”, Arxiv Draft, 2021
[2] “Toward Personalized Emotion Recognition: A Face Recognition Based Attention Method for Facial Emotion Recognition”, Accepted to IEEE International Conference on Automatic Face and Gesture Recognition, 2021
[3] “Driver Drowsiness Detection”, Patent ID 87208558, Filed in 2021}

Research Experience:
Capstone Research Student at University of Toronto September, 2021 ~ April 2022
<sub>- Worked on a research project about adding new personalization and communication efficiency features regarding Hierarchical Federated Learning (HFL) to FedML, an open-source Federated Learning library.
- Applied FedPer to achieve personalization in client level and edge level. Designed a CNN-based layers division strategy and studied the influence of the sampling portion, and achieved a 12% accuracy improvement. Utilized dynamic sampling, probabilistic selection scheme and dithered quantization to reduce data transmission by 77.5%.
- Quantitatively and qualitatively tested on the Federated-EMNIST dataset and the UTKFace dataset.
- The project won a certificate of distinction.</sub>
Research Assistant in D3M Research Group at University of Toronto May ~ September 2021
<sub>- Worked on an explaining-driven data augmentation method to improve faithfulness between image classification models and explanation method together with the LG-AI research group and contributed to a paper [1].
- Participated in experiments design and implementation; investigated and conducted an improved qualitative and quantitative analysis; tuned hyper-parameters and refactored the codebase to reduce cyclomatic complexity; designed automatic batched evaluation methods.
- Evaluated 2 models (Resnet50 and VGG16) on 2 datasets (CIFAR-100 and PASCAL VOC 2012) and improved explaining metrics including drop & increase, insertion & deletion, ebpg & miou, etc and outperformed 5 classical baseline methods (vanilla training, cutmix, mixup, augmix and random explanation).</sub>

Internship Experience:
Assistant Computer Vision Engineer in Noah’s Ark Lab, Huawei Canada May, 2020 ~ April, 2021
<sub>- Designed a new driver drowsiness detection application with MTCNN and CNN based backbone models and implemented both the Python prototype and Android final product. Contributed to a paper [2] and a patent [3].
- Investigated open source face detection and facial landmark localization algorithms; evaluated and benchmarked each approach under the specific settings (model size limit, inference time limit, robustness, various people appearances).
- Implemented data processing and hyper-parameter tuning; refactored code to increase readability and reliability; conducted system tests for acceleration and inspected error reports from the Beta users to fix emerging defects.
- Improved CV capability metrics: eye close per frame precision from 80% to 88% and recall from 60% to 94%; improved speed from 5fps to 15fps by code refactoring and industrial tricks.
- The final product, Driving Care, launched on the Huawei HiCar device in 2021.</sub>
Data Analyst Intern in Agriculture Bank of China July ~ August, 2019
<sub>- Participated in a data mining competition about loan overdue predictions and ranked top 10% among the participants.
- Developed models using Python (NumPy, Pandas) for data preprocessing and feature engineering.
- Increased prediction KS value to 57% by optimizing algorithms and tuning parameters.</sub>

Project Experience:
File Transfer Via Self-Designed Protocol in C March 2022
<sub>- Used Unix socket programming to implement a server and client program which conducted file transfer with UDP protocol.
- Embedded information including fragment number and size into packet strings; designed a stop & wait acknowledgement strategy.</sub>
Caching, Multi-threaded Web Server in C December 2021
<sub>- Designed a web server to handle requests from multiple clients; used Mutex Locks and Conditional Variables for synchronization.
- Used a hash table as the LRU-cache with maximum limit to utilize the locality of reference for acceleration.</sub>
Dive Booking Application Database Design April 2020
<sub>- Studied the constraints requirements and designed the database schema in DDL with optimization in redundancy reduction.
- Populated comprehensive instances of data into the database and wrote SQL queries to solve simulated real life problems.</sub>
Traffic Sign Detection project April 2020
<sub>- Enriched size of dataset by utilizing data augmentation based on Python to create training and validation sets.
- Implemented color and shape detection models to identify and localize traffic signs in the sample image data.
- Classified dataset into two stages and applied machine learning techniques including 7-layer CNNs to process data in each stage; reached overall 60% accuracy of traffic sign recognition and localization in a driving record.</sub>
C++ Map project April 2019
<sub>- Developed C++ based algorithms for a full stack map application; designed and completed frontend user interface with ezGL and GTK.
- Implemented a navigation system using A* algorithm and Dijkstra to solve the shortest path issues.
- Integrated and combined greedy algorithm, simulated annealing, 2-opt, and 3-opt algorithms, genetic algorithms, etc., to solve core features related to the salesman problem.</sub>

Ongoing Personal Projects:
Leetcode Problems Daily Practice Since July, 2020
OpenCV Cube Solver (Reproduction of a YouTube Video) Since February, 2021
<sub>- Find contours and detect color in regions with OpenCV library.
- Solve the cube with the kociemba module.
- Will upload the open source code when I finish it.</sub>