Po-Lin Chen Email & Phone Number

台灣 臺北市

Advisor: Prof. Ta-Te LinTopic: An Efficient AIoT Framework for Video-based Behavior Monitoring in Dairy Calves• Identified MoViNet-Stream-A0 with Float16 Quantization as the optimal settings for real-time video analysis in dairy calves through performance tests on Raspberry Pi system, achieving 47ms latency and a 0.916 F1-score• Devised and deployed a computation-efficient AIoT framework that integrated on-device inference with cloud services, demonstrating its effectiveness in real-world settings through long-term trials at NTU’s Experimental Dairy Farm

台灣 臺北市

Introduction to Artificial Intelligence (Fall 2024)• Guided AI-focused Python projects for 60+ students involving ImageBind for multimodal representation learning, NeRF for 3D scene reconstruction, FastSpeech for audio synthesis, and GPT for language tasksComputer Programming Language (Fall 2024)• Directed 2-hour C++ lab sessions and office hours weekly to support 60+ students throughout 10 assignmentsPractical Data Structures and Algorithms (Spring 2024)• Formulated application-oriented coursework on fundamental data structures and algorithms in Java• Established and maintained an integrated online judging system to enable automated grading for 110+ students

台灣 臺北市

Advisor: Prof. Ta-Te LinTopic: Transformer-based Dairy Calf Action Recognition and Analysis• Leveraged Video Masked Autoencoder (VideoMAE) to recognize 7 behaviors in dairy calves from video frames• Boosted overall accuracy by 3.7% through the use of bidirectional LSTM to rectify behavioral sequences• Identified 2 strong behavioral indicators to Neonatal Calf Diarrhea through feature importance and statistical analysis

台灣 新北市

Topic: Edge Vision AI Model Development and Product Integration• Enhanced existing self-checkout kiosks with an advanced multi-modal alarm system that fused vision-based anomaly detection and barcode signals, driving data-driven loss prevention and labor reduction in retail• Optimized stateful CNN models through Intel’s OpenVINO C++ runtime and efficient state buffer management, achieving 2.7X faster inference on industrial PCs with Intel CPUs and iGPUs• Minimized false alarm rate to 2% while maintaining 99% precision in detecting 4 unusual behaviors