Yichen Li Email & Phone Number

Guangzhou, Guangdong Province, Cn

Santa Clara, Ca, Us

Southfield, Michigan, Us

➢ Implement Python project to generate Signal Phase and Timing (SPaT) data using Traffic Signal Controller (TSC) data live-streaming from both WebSocket and AWS S3 bucket, which will replace current SPaT monitor components in TSC to replace all SPaT units and will provide backend support to team’s current web application Way-Logic with integration to Google map➢ Implementing a traffic light optimization pipeline based on Federated Deep Learning (FDL) with an A2C model, using Flower framework, with each intersection belonging to a local training machine, using Horizontal FL to build a network-level optimization model.➢ Improving the company's current prediction algorithm (Step Attention) for multiple probabilistic trajectory prediction using GMM. Also, introduces Geo-fence to constrain the prediction results by aggregating features with Geo-information using Cross Attention.

Ann Arbor, Michigan, Us

Grader of EECS 498-010, Principle of Machine Learning

San Francisco, Ca, Us

➢ Implemented an end-to-end software system log analysis pipeline using Python, including data ETL, data feature extraction based on log event type and variables vectorization, and a log event time-series sequence prediction method based on the multivariant-Bi-LSTM algorithm. The algorithm can maintain the precision of more than 96% of diagnosing abnormal log messages. It will be integrated into the NFV system, an essential layer in a communication system, to prevent critical software failures and help developers determine workflow➢ Collaborated on developing and training a GAN generator for log data generation to compensate for the lack of an existing data set.➢ Worked on using K-anonymity and Differential Privacy methods to build an anonymization tool to keep the privacy of existing log data

Santa Clara, California, Us

➢ Implemented an end-to-end Machine Learning pipeline, using datasets with sizes more than 10TB, using the recommendation model, Deep Interest Evolution Network (DIEN), using NumPy, Pandas, TensorFlow, and Modin, making the performance of data ETL and feature extraction parts 10 times better in single-threaded processing, and deployable in distributed systems➢ Collaborated with the full-aspects analysis and optimization (application, framework, and cluster) of machine learning and big data projects deployed on Hadoop, Hive, and Spark for different workloads, including SQL queries, MapReduce, and multiple machine learning tasks, according to the performance statistics from benchmark tools TPCx-BB & TPCx-DS, providing critical suggestions on optimization tasks to the team, and rewritten the company’s guide book of Hadoop and Spark configuration for multiple tasks

Pittsburgh, Pa, Us

➢ Implemented APIs for Document for Excel, one of GrapeCity’s flagship software products using C# and Java, with functions like cellpadding, cross-file references, JSON IO with options, etc. No major bug reports after each release➢ Collaborated on developing basic package function Calculation Engine, implementing data manipulation logic

Hubei, Cn

Duty: Help develop and program for part of the software that will be integrated to PET machine. Use machine learning techniques and MVT theory to process PMT signal sampled from PETDescription: Traditional ADC devices cannot be integrated on PET due to their large size, but small-volume samplers cannot be used directly because of the low sampling rate. Thus, it needs further signal process to make up for the information loss.

Graduate Certificate, Data Science

Master Of Engineering - Meng, Data Science And Machine Learning

Summer Program, Information Science And Engineering

Bachelor Of Engineering - Be, Artificial Intelligence And Automation