Activities / achievements other than taking a career break* 🇩🇪 Learned German: - Took German B2 intensive courses - Took "Diskutieren und Präsentieren auf Deutsch; B2-C1" course at UZH - Passed TELC B2 exam with "Sehr Gut" level achieved* 🥌⛷️ Learned new sports: Curling and skiing * 🗣️ Learned basic LLM: "Generative AI with LLMs" on DeepLearning.ai* 👩‍🎓 Completed Capstone Project and obtained DAS ETH in Data Science diploma* 👩‍💻 On-going side project "RL2048": https://github.com/FangLinHe/RL2048

6DOF object detection and tracking.

Captural merged with Ifolor Zürich AG in August, 2021. Continuing work on Machine Learning and Computer Vision topics. Continued developing our cross-platform CVML library and made PoC on Windows, Android, and iOS platforms.

Categorizing and searching photo libraries and smart photo editing:• Designed and implemented ML data pipeline, CV & ML algorithms and models for object and scene classification, face clustering, and recognition of topics and protagonists in photo albums (Python, PyTorch, scikit-learn, Google Cloud Platform, Apache Beam)• Ported CV & ML algorithms for on-device execution and developed iOS and Android prototype apps (Swift, Objective-C, CoreML, ONNX, Java, TFLite)• Productized CV & ML algorithms in the form of cross-platform library: API design, automatic generation of language bindings, CI/CD pipeline with 95% test coverage and automatic releases (Modern C++, Azure DevOps)

Scene understanding with Deep Learning for smart signage solutions• Collaborated with ETH Zurich to develop multi-task-learning model for bodyskeleton, face, age and gender detection (Python, PyTorch)• Optimized body skeleton model inference time by 78% without affectingaccuracy (Python, TensorFlow, TensorRT)• Converted Deep Learning models from TensorFlow, MxNet, PyTorch to ONNXfor deployment; deployed models with TensorRT (Python, ONNX, TensorRT)• Implemented incremental DBSCAN clustering to speed up face cluster updateby 1 order of magnitude; implemented method for tracking multiple people jointly with two cameras (Modern C++)

• Designed DL model for real-time age estimation with average error of 6 years and far-distance gaze direction estimation with average error of 30° (Caffe)• Developed musical game controlled with facial expressions detected by ML model (Modern C++, Qt)

Built RCNN-based object detector to detect and recognize smart phone devices: labelling tool, data preprocessing, model training and deployment. (Python, Caffe)

• Planned, designed, and implemented OEM project "Virtual Design Center" to layout furniture in real world using augmented reality technology, running on Windows, Android & iOS platforms (C#, Unity)• Gave several education training sessions to universities students, staffs, and professors; topics including Unity 3D, hardware installation and usage of various products