○ Improved the accuracy of auto-label engine, and the proposed an evaluation approach. ○ Performed social-distance detection project.

○ Developed ground-truth level auto-labeling algorithm for object detection, e.g., face, human, animals, vehicle. ○ Studied auto-ml for edge Ai

○ Researched solutions for model compression for ADAS.

○ Initiated turnover prediction project for direct labors.

○ Studied machine learning algorithms and Apache Spark's core includes mapreduce and RDD's cache, partitioning and optimization, dataframe, and mlib. ○ Simulated channel assignment using Spark.

○ Built a research prototype, including a data gathering and analysis environment based on PTT the largest BBS in Taiwan, to understand from counseling point of view whether online wisdom of crowd can be applied for professional helping service. ○ Built a new crowdsourced incentive Q/A platform.

○ Led a team 4 computer science graduates to design a fast transcoding from H.264/AVC to H.264/SVC, a fast mode decision to speed up H.265/HEVC and a low complexity rate model of H.264 CABAC accepted by IEEE VCIP14 .

○ Designed a rate model and controller for H.263 in TMN8 IEEE ICASSP06, IEEE ICC07, IEEE CSVT08, IEEE ISJ2 012 ○ Designed a low complexity rate mode of H.264 CAVLC in JM IEEE ICASSP 07 . ○ Led a team 6 graduates to prototype a live p2p streaming system. ○ Conducted an extensive event-driven simulations in P2PSim, and deduced a mathematical model to evaluate bandwidth utilization when some high-priority information is delivered in a swarm fashion so that we can optimize the delivered video quality in swarm-based p2p streaming system SIGCOMM09 poster, IEEE ICME09 oral, IEEE ICNP1 1 poster, IEEE HPCC11.