Tsung-Shan (Kevin) Yang
Ph.D.(Defense passed) in ECE, University of Southern California
About Me
Tsung-Shan (Kevin) Yang is a Ph.D.(Defense passed) in Electrical and Computer Engineering at the University of Southern California (USC), advised by Professor C.-C. Jay Kuo. He received his Bachelor’s degrees in Chemistry and Electrical Engineering from National Taiwan University (NTU) in 2019, and his Master’s degree in Electrical Engineering from NTU in 2021.
His research centers on image understanding and machine learning, with a particular focus on developing efficient and interpretable learning frameworks. His work emphasizes principled model design and explainability to improve robustness and practical deployment.
His publications and ongoing research can be found on Google Scholar.
Education
Adivisor: Professor C.-C. Jay Kuo
Thesis: Interpretable and Efficient Multi-Modal Data Interplay: Algorithms and Applications
GPA: 3.81/4.30
Dean's List for 2 semesters (2014 Fall, 2015 Spring)
Work Experience
Achieved state-of-the-art performance with only 3% of model parameters and a 98% reduction in inference time.
Journal Publications
HOI Detection / Image Understanding / Green Learning
Efficient human–object-interaction (EHOI) detection via interaction label coding and Conditional Decision
Computer Vision and Image Understanding (2025)
A lightweight and human sensible human-object interaction detector.
Multimodal Alignment / Image Understanding / Green Learning
Image-Text Retrieval via Green Explainable Multi-modal Alignment (GEMMA)
APSIPA Transactions on Signal and Information Processing (2025)
Explainable alignment for features from two separately trained models in different modalities (image and text).
Omnidirectional Video / Saliency Detection / Center Bias
Viewing Bias Matters in 360 Videos Visual Saliency Prediction
IEEE Access (2023)
Statistically analyze the human-bias in saliency maps and generalize the spherical kernel to time series data.
Conference Publications
Video Understanding / Generated Content Detection / Efficient AI
SVD-Det: A Lightweight Framework for Video Forgery Detection Using Semantic and Visual Defect Cues (WACV' 26)
97% smaller, 98% faster, while achieving +2.7% higher AUC for AIGC video forgery detection.
Multimodal Alignment / Video Understanding / Green Learning
Interpretable Video-Text Alignment (VTA) for Cross-Modal Retrieval (APSIPA ASC' 25)
Deploy an explainable ranking pipeline for video-text retrieval, reducing 97% trainable parameters.
Multimodal Alignment / Image Understanding / Green Learning
GMA: Green Multi-Modal Alignment for Image-Text Retrieval (APSIPA ASC' 24)
Explainable alignment for features from two separately trained models in different modalities (image and text).
Human-Object Interaction Detection / Image Understanding / Green Learning
GHOI: A Green Human-Object-Interaction Detector (IEEE MIPR' 24)
A lightweight and human sensible human-object interaction detector.
Point Cloud / Quality Assessment / Green Learning
BPQA: A Blind Point Cloud Quality Assessment Method (IEEE ICIP' 23)
A point cloud quality assessment with statistical feature extraction and interpretable learning scheme.
Dehazing / Real World Image / Deep Learning
NTIRE 2020 Challenge on NonHomogeneous Dehazing (IEEE CVPR' 20)
Propose an attention refinement block of the deep learning model.
Few Shot Learning / Clustering / Deep Learning
Few Shot Learning With Difficult Setting (CVGIP' 18)
Analyze the different approaches of few shot learning.
IC design / Machine Learning
IR Drop Prediction of ECO-Revised Circuits Using Machine Learning (IEEE VTS' 18)
Reduce 30X simulation time through deep learning.
Teaching Experience
Introduction to the basic c++ syntax and algorithms.
Design deep learning projects and assignments.
Maintain the course website.
Provide 2 hour TA class each week for over 300 students.