Master of science & Computer Engineering

2018 - 2020

The Fu Foundation School of Engineering and Applied Science, Columbia Univesity in the City of New York, NY

Major Coursework: Applied Deep Learning, Deep Learning for Computer Vision, Artificial Intelligence, Computer Architecture, Analysis of Algorithms, Databases, Datacenter Networks, Formal Verification of HW&SW, Internet of Things - Systems and Physical Data Analytics

Bachelor of Engineering & Electrical and Computer Engineering

2014 - 2018

Computer Science and Control Engineering, Nankai University, Tianjin

Major Coursework: C++, Data Structures, Machine Intelligence, Principles of Computer Organization, Visual Program in Control System, Sensing Technology and Application, Automatic Detection Technology & System

AutoML using Metadata Language Embeddings

2019

Columbia University, New York, NY

Objective: provide a zero-shot AutoML system using metadata embeddings and pipeline embeddings.

• Applied USE3 embeddings from modern NLP to the dataset description and algorithm documentation and learnt the non-linear interactions between these embeddings using a designed neural network to get the recommended pipeline of the most similar dataset.
• Demonstrated our AutoML system’s performance running in 1s improved on the performance of state of art AutoML frameworks AutoSklearn, AlphaD3M, et, al. in 1 min based on our designed execution engine.

Design and Implementation of an Immersive Task for Rehabilitation

2018

Nankai University, Tianjin, CN

• Established an immersive VR rehabilitation system based on Oculus Rift VR platform.
• Designed the bilateral arm training tasks and proposed the evaluation index;
• Conducted a series of experiments based on the designed tasks, including the unilateral and bilateral comparison experiments and the verification experiments of learning effect.
• Built a non-immersive VR rehabilitation system, separately conducted experiments in the two rehabilitation systems based on the designed tasks, and analyzed the experimental results quantitatively.

Target Identification and Autonomous Tracking for Vision-based Quadrotor UAVs

2016 - 2018

Nankai University, Tianjin, CN

Objective: make UAV identify and autonomously follow the given target without GPS in an unknown environment.

• Utilized TLD (Tracking-Learning-Detection) algorithm using P-N learning estimating detector’s errors to long-termly track the target in real-time based on the video stream obtained from Kinect V2.
• Developed an algorithm for SLAM using C++: extracted feature points with SIFT algorithm from the RGB-D data, computed the rotation matrix with the quaternion, and estimated the pose of the vehicle using PnP algorithm.
• Constructed the 3D point cloud map of the environment.
• Won the First Prize for Undergraduate Innovation Project.

Teaching Assistant of COMS 4995 Deep Learning course

2019

Columbia University, New York, NY

• Gave tutorials on TensorFlow, Keras, TF-GAN, and TF-Agents [TensorFlow, Keras, TF-GAN, TF-Agents].
• Helped the instructor and students with tasks such as grading, office hours, etc.
• Mentored several research projects in cutting-edge topics in DL.

Detection and Classification of Breast Cancer in Whole-Slide Images [Grand Challenge]

2019

Columbia University, New York, NY

Objective: develop a tool to assist physicians to locate the breast cancer in slide images.

• Developed a multi-scale approach using Inception V3 architecture as the pretrained model that utilize patches extracted by a sliding window at multiple magnifications centered on the same region.
• Trained and evaluated the designed CNN framework on Google Cloud Platform using TensorFlow on the large and tumor class imbalanced CAMELYON16 dataset (around 3TB) and achieved a recall score of 98%.

6D Pose Prediction of Vehicles in Autonomous Driving [Kaggle Competition]

2019

Columbia University, New York, NY

Objective: predict the 6D pose (position and rotation) of each vehicle from monocular RGB image.

• Implemented an anchor-free method using PyTorch by designing a convolutional network based on UNet and FPN (Feature Pyramid Network) with ResNet as the backbone, trained on over 4200 images, and ranked top 20% in this competition.
• Applied Gaussian kernel to generate ground-truth heatmap and used CoordConv to avoid space invariance feature of CNN.

Smart Home Monitor

2019

Columbia University, New York, NY

Objective: provide real-time monitoring of home condition, and improve the home security with face recognition remotely.

• Built a home monitor website using HTML, JavaScript, CSS, displaying the dynamic light, gas, et al. monitor in real-time with ICN (Information-Centric Network).
• Constructed a serverless backend using AWS services including DynamoDB, S3, EC2, Rekognition API, Lambda function, and SNS; fulfilled real-time embedded face recognition using TensorFlow and AWS S3.
• Used Ajax to exchange the data between the server and the client based on the RESTful API with Flask and AWS DynamoDB. [HTML]

Database Centric Web Application Focusing on Baseball Data

2019

Columbia University, New York, NY

Implemented the following microservices:

• Baseball Data Microservice: Allows users to query and explore data about players and their performances; allows users to create an account, assign players to teams, etc. with a newly designed relational database and data model. (CRUD with MySQL)
• Social Interaction Microservice: Allows users to follow, like, comment, etc. on teams, players, users, based on Neo4J.Social Interaction Microservice: Allows users to follow, like, comment, etc. on teams, players, users, based on Neo4J.
• Discussion/Comment Data Microservice: Stores comments, responses, discussion threads, etc. based on AWS DynamoDB.
• Application Business/Logic Microservice: Implement business logic and rules. (REST API backend based on Flask framework)
• Caching Microservice: REST API and data access response cache to optimize performance, based on Redis.