Heiru Wu Email & Phone Number

Phoenix, Arizona, United States

• Designed PoC Machine Learning solution which combines optical flow feature characterization and color-based image classification DNN model for fire hazard detection in large scale semiconductor fabrication plant.• Deployed on complex infrastructure with High Availability virtual machines migration system to acheive the ability to inference 200+ real-time video streams with 0 downtime.

Hsinchu City, Taiwan, Taiwan

• Designed and developed a Machine Learning enabled home-caring system which introduced intelligent ambient awareness powered by our modularized inference components, delivering features that prevents 90% or more of the accidents and continuously adapting to current environment and new behaviors.• Developed custom MLOps framework for automation pipeline to perform self-annotation, active-learning, performance evaluation and model deployment to cut down 90% of the human resources and maintain 99% of system availability and reliability.

Hsinchu City, Taiwan, Taiwan

• Designed and developed a Machine Learning enabled home caring system which introduced intelligent ambient awareness powered by our modularized inference components , delivering features that prevents90% or more of the accidents and continuously adapting to current environment and new behaviors• Developed custom MLOps framework for automation pipeline to perform self annotation, active learning,performance evaluation and model deployment to cut down 90% of the human resources and maintain99% of system availability and reliability• Researched and developed modularized Machin Learning inference components including object detection, object tracking, pose-estimation, action-recognition and person re-identification, which can speed up task-specific domain adaptation up to 80% and cut down 50% of development time.

Hsinchu County/City, Taiwan

Customized deep learning backbone networks including Mobilenet, Darknet 53 and CSPNet,combined with SSD and Yolov3/v4 detection network and object tracking algorithms, deployed a n objectdetection system for self driving vehicles with above 9 0 % intra class averaged mAP and f1 score

Columbia, South Carolina Area

Redesigned an open source deep learning steering angle prediction model built with Tensorflowframework, we introduced novel convolutional blocks to learn and extract additional high dimensionalfeatures with 20% faster model forward time and 10 % higher mAP over custo m dataset

Developed tools to help construct 3D pointcloud vector map for autonomous vehicle ’s SLAM algorithmand route planning while utilizing PPP GPS to correct the margin of error to achieve 90% accuracy

Hsinchu County/City, Taiwan

3D human model reconstruction and body scale measurement• Designed image recognition algorithms and developed an application that uses Xbox Kinect SDKv2, implementing the algorithms to extract raw data then process into pointcloud information for 3D human model reconstruction and contour detection for 2D-to-3D model scale estimation.• This project uses only one camera instead of conventional three or more cameras system, while maintaining 98% of accuracy; also, with the reduce amount of raw data and advance 2D-to-3D conversion algorithms, the speed of processing data is up to 3X compare to conventional model.

195, Sec. 4, Chung Hsing Rd., Chutung, Hsinchu, Taiwan 31040, R.O.C.

Tomato project, an IFTTT-like web-based service• Developed a user-friendly Android app for the service, with well-designed UI hooking the database and web APIs with SQL and http commands respectively, letting users easily and effortless adjust their preference with the scripts on the go. The service is integrated with 80% of the mainstream social media services and various IoT services like Philip hue.• The app and server are able to handle over 10,000 users online simultaneously by pressure testing. Hardware acceleration codec implementation on Android system• Implemented OpenMAX open-source codec library to process real-time video streaming with front camera, which supports H.264 with roughly 2X speed-up compare to built-in software decoding speed.

Chang Gung University

Development of a Wearable Multi-Posture Detection Body Sensor Network• Designed a 16-bit floating point data packet structure that is used to wrap the three-dimensional coordinate information from the accelerometers. And developed a monitoring system on Android platform with integrated algorithm for analyzing data which detects dangerous posture by patients.

Chang Gung University

Development of Four-Wheel LEV with integrated power chain and driving monitoring system• Developed a dashboard system on Android platform that interact with the microprocessor on the LEV under BLE protocol, which provide critical information of the vehicle, including speed, acceleration, battery level and processor temperature. The system is integrated with google map APIs, open source codec for dash-cam recording, speed limit voice warning and security system that prevent any risks.

Master Of Science - Ms, Computer Software Engineering, 3.9

Bachelor Of Science (B.S.), Electrical And Electronics Engineering