Software Engineer Intern,   05. 2019 - 08. 2019

       Improved the clustering module in testing system to detect duplicated bugs by 7 percents in F-score.

       - Optimized feature engineering to apply a clustering service on the product system.

       - Extended similarity measure options and refactored related code to unify the interface.

       - Added the clustering prediction and evaluation binary interfaces.

       - Designed a pre-evalutaion metric when the dataset is limited and analyzed algorithms performance.

       Deep Learning Intern,   04. 2018 - 07. 2018

       Contributed to 0.1.0 and 0.2.0 releases and 4 PoCs of Analytics Zoo, an open-source Analytics + AI Platform for Apache Spark and BigDL.

       - Anomaly Detection: Implemented RNN model (with LSTM layers) to detect anomaly temperature and reduced model memory from 22g to 8g by feature engineering and model modification.

       - Object Detection: Used SSD & F-RCNN models to detect instances of certain classes in images and videos.

       - Image Augmentation: Developed 16 ways to transform images and designed transformation classes using polymorphism to enable sequential operations.

       - Wrapped Scala APIs to Python APIs and built jupyter notebook examples.

       Software Development Engineer Intern,   02. 2017 - 05. 2017

       Work on "FILEGEAR", a private cloud service that supports cloud file management and sharing.

       - Built an independent function model for iOS client using Swift and completed UI renovation to target iOS users.

       - Constructed server’s APIs to enable the development of new features.

       - Implemented up to date models including TATransE, TADistMult and TTransE to reproduce their stated results. [Code]

       - Built data preprocessing utils and analyzed datasets statistics.

       - Experimented the proposed model and baseline models on four datasets for comparison.

       Publication: "Recurrent Event Network for Reasoning over Temporal Knowledge Graphs”. Woojeong Jin, Changlin Zhang, Pedro Szekely, Xiang Ren. ICLR 2019 Workshop on Representation Learning on Graphs and Manifolds.

       Aim to improve accuracy of semantic segmentation on mobile device with limited computing resource.

       - Applied innovative Squeeze-and-Excitation on baseline model of real-time semantic segmentation.

       - Provided the workflow of image augmentation and pre-processing the newest released dataset ADE20K.

       - Experimented improved model and baseline model on indoor dataset, outdoor dataset and mixed dataset to analyze results.

       An android application to recognize items on receipts and do data analysis in order to help with people’s financial management.

       - Applied Tesseract Open Source OCR Engine and its training data to recognize items on receipts.

       - Classified items into different categories according to names by Deterministic Finite Automation (DFA) to achieve data analysis.

       - Built an Android application utilizing a user-friendly design to support industrial usage.

       An advanced search engine over Reuters Corpus to satisfy non-professional users.

       - Implemented boolean model and vector space model (based on tf-idf consine similarity).

       - Added advanced features (e.g. topK query, phrase query, synonym query, wildcard query and spelling correction).

       - Simulated prevailing search engines with well-designed user interface.

       A search engine over "The Works of William Shakespeare"

       - Identified the stop words with universal stop table and statistical way for specific Shakespeare set.

       - Optimized the thresholds on query to achieve optimal results.

       A news aggregator from mainstream news media websites regularly for UI integration.

       - Rearranged UI presentation to learn and adapt through users’ behaviors for specific customization.

       - Adapted the responsive layout to support both web and mobile clients.

       Concise SQL engine (Database Management System) for single user to create and delete tables, indexes, records and select records thorough a character interface.

       - Implemented a buffer manager module with Least Recently Used (LRU) and pin function to minimize disk read.

       - Implemented an interpreter module and designed the flow of control and parse tree.

       Employer: Professor Yin Zhang,   DCD lab

       - Tutored 112 students in the weekly lab session in addition to daily office hours.

       - Adapted programming assignments to ensure the students to comprehend the course materials.

       - Set evaluation standards for grading, graded homework with valuable feedback and reported to the professor existing problems with causal analysis.