PyTorch Inference enablement

Localization & mapping team: Machine Learning Mapping

• Open Neural Network Exchange (ONNX): Grew the ONNX Community by providing new features, solving issues, improving existing functionalities. (Python, GitHub)

• Release: Drive ONNX 1.8.0 Release on various platforms as a Release Manager. Fix platform issues and make it support backward and forward compatibility. (PyPi, conda, GitHub Action, Travis CI)

• Communication: Intensively cooperated with other teams.(ONNX Runtime, Pytorch, Tensorflow, Caffe2, MLIR)

• Shape inference: Improved Deep Learning operators and implemented ONNX graph checker. (C/C++)

• Enhance Robustness: Developed and maintained CI on various platforms.(Azure Pipelines, CircleCI, Travis)

https://github.com/onnx/onnx

• Open Neural Network Exchange: Utilized ONNX Runtime for performance tuning among 6 deep learning

frameworks. Applied 30+ models on ONNX pipeline, investigated them and minimized the errors.

• Notebook Interface: Streamlined complicated Docker commands on Azure into simplified Python library. (10+ commands into 2 commands) Designed comprehensive interface and refined visualization on Jupyter Notebook.

• Improve Usability: Strengthen usability for ONNX by writing complete automated testing codes to handle any exception. (unittest) Designed a user-friendly and interactive Web platform for non-technical users.

Github: https://github.com/microsoft/OLive

• Developed Scientific Knowledge Graph supervised by Prof. Jia Zhang.

• Scientific Knowledge Graph: Upgraded a retrieval system (Java & Play) for scientists in NASA Research.

• Enhance Robustness: Expanded a backup system (MySQL) to sync 8000+ papers from the remote database (MongoDB) and designed a daily updating scheduler; Solved 20+ issues per week. (top 1 in our lab by JIRA)