Professional Experience

There is no doubt that each internship and research opportunity has brought me different aspects of harvest and growth. I would like to thank all the professors, Ph.D./MPhil candidates, mentors, and colleagues for their generous help and guidance.

INTERNSHIPS

Covariant(Shenzhen)

Machine Learning Engineer Intern in Shenzhen, China
May. 2022 - Jul. 2022 | Mentor: Yide Shentu

Implemented relevant structures and classes with the Mechanical team to make a new type of small USB wrist camera compatible with the company-wise video recording & data collection pipeline
Wrote scripts to let robot gradually gain speed in the picking cycle for the same object until it falls, which automated the long-tail data collection process, and ran smoothly for more than 20 hours
Constructed a self-supervised video segmentation pipeline with 3D vision and rendering techniques to predict the mechanical arm silhouettes, which could help to detect robot picking failures in a third-person view

Hikvision Digital Technology

Software Engineer Intern in Hangzhou, China
Jun. 2020 - Aug. 2020 | Mentor: Yishi Deng

Processed and analyzed thousands of lines of Excel & Html records of Project Subunit information and organized them into the dataset
Generated and visualized Project Subunit dependency tree for project planning and evaluation with Java and SQL

RESEARCH PROJECTS

Online Easy Example Mining for Weakly-supervised Gland Segmentation from Histology Images

Undergraduate Research Assistant in HKUST, Hong Kong
Sep. 2021 - Mar. 2022 | Supervisor: Prof. Xiaomeng Li
MICCAI 2022 Accepted Paper

Analyzed and pointed out that the key problem of gland segmentation is the confusion caused by the morphological homogeneity of histology images, rather than the local activation problem that most WSSS methods faced
Proposed the Online Easy Example Mining (OEEM) technique to mine the credible supervision signals in pseudo-mask, mitigating the damage of confused supervisions for gland segmentation
Designed a robust two-stage framework for weakly-supervised gland segmentation, achieving a State-of-the-arts mIoU result of 77.56% on the GlaS dataset with the paper published by MICCAI 2022

Undergraduate Research Assistant in HKUST, Hong Kong
Jan. – Aug. 2021 | Supervisor: Prof. Xiaojuan Ma
DIS 2022 Accepted Paper [Appears in Acknowledgement]

Proposed a template-based self-mockery generation method, which could construct four related components in the dialogue context and fit in the pre-defined template for real-time human-robot interaction
Built the Self-mockery Robot with the RASA platform and deployed it on the web for subsequent User Study
Studied the processing methods and dialogue of robots on several online shopping platforms and designed the baseline chatbot language without self-mockery function as the control group for comparative experiment
Participated in statistical results analysis and published our work on DIS 2022, which pointed out that the self-mockery design significantly improved chatbot’s damage control ability and emotional intelligence

Lung Tumor Segmentation and Drug Resistance Prediction Through Novel Deep Learning Architectures

Undergraduate Research Assistant in HKUST, Hong Kong
Jun. – Aug. 2021 | Supervisor: Prof. Xiaomeng Li

Cropped out the lung part and eliminated background noise of the 3D CT lung images from Queen Mary Hospital
Utilized Multi-Modality learning method with nnUNet to train a segmentation model for lung tumors and achieved over 80% mIoU score
Appended a classification branch at the lowest level of the segmentation network for the Drug Resistance Level prediction, iteratively frozen and trained the two branches, and got over 70% classification accuracy

Yiduo YU 余亦多