CV
Contact Information
| Name | Ting Yu Tsai |
| ttsai2@albany.edu / tiffany9056@gmail.com | |
| Work Authorization | No sponsorship required; permanent resident |
Summary
Ph.D. candidate in Computer Science with experience building multimodal LLM, machine learning, computer vision, and Python-based software systems for real-world applications. Developed end-to-end deep learning pipelines, evaluation workflows, and full-stack AI tools for medical imaging, video understanding, and intelligent decision-support systems. Brings strong hands-on experience in Python, PyTorch, TensorFlow, multimodal LLM inference, model evaluation, and software development, with a track record of translating research ideas into practical and scalable solutions.
Education
Skills
Projects
-
ReDiPrune: Efficient Multimodal LLM Inference
A training-free token pruning framework for multimodal LLMs.
- Developed a training-free token pruning framework for multimodal LLMs that improves inference efficiency by pruning visual tokens before the vision-language projector.
- Designed a lightweight relevance-diversity scoring method to preserve query-relevant and non-redundant information.
- Enabled plug-and-play integration into existing multimodal LLM pipelines without retraining or architectural changes.
-
IntelliCardiac: End-to-End AI Platform for Clinical Decision Support
A full-stack AI platform for cardiac MRI analysis and disease classification.
- Built an end-to-end AI platform for cardiac MRI segmentation, five-class disease classification, and real-time result visualization.
- Developed model pipelines, evaluation workflows, and deployment-oriented application components using Python, PyTorch, JavaScript, Node.js, and MongoDB.
- Achieved 92.6% overall segmentation accuracy and 98% classification accuracy in the two-stage disease classification pipeline.
-
Rat Seizure Video Analysis Framework
A deep learning and computer vision framework for automated seizure analysis from video.
- Developed a video analysis pipeline for automated rat seizure detection and stage classification.
- Built a scalable and reproducible framework for behavioral monitoring in epilepsy research using deep learning and computer vision.
-
UU-Mamba: Robust Cardiovascular Image Segmentation
A cardiovascular image segmentation framework designed for robustness and generalization.
- Developed and benchmarked a segmentation framework integrating Sharpness-Aware Minimization and an uncertainty-aware composite loss.
- Improved robustness and generalization for cardiovascular image segmentation across multiple medical imaging datasets.
- Achieved state-of-the-art performance on the ACDC, ImageCAS, and Aorta datasets.
-
Automatic Image Exposure Classification and Enhancement System
A deep learning-based system for image exposure classification and enhancement.
- Built a deep learning system for automatic image exposure classification and enhancement.
- Achieved 0.91 classification accuracy and improved image quality to a level comparable to camera-captured results.
Experience
-
2025 - 2025 Albany, NY, USA
Research Assistant
University at Albany - SUNY
Mamba-2 Quantization and Model Evaluation
- Adapted the Quamba2 pipeline for Mamba-2 model quantization and reproduced benchmark results across model scales, enabling systematic evaluation of inference efficiency under quantization.
- Built evaluation workflows to profile latency and memory usage and analyze activation behavior, state-space dynamics, and performance trade-offs across model variants.
-
2024 - 2024 Albany, NY, USA
Research Assistant
University at Albany-SUNY
Cardiovascular Image Segmentation
- Developed and benchmarked Python/PyTorch deep learning pipelines for cardiovascular image segmentation across challenging medical imaging datasets.
- Integrated Sharpness-Aware Minimization (SAM) and uncertainty-aware loss design to improve model robustness and generalization.
- Designed and executed systematic evaluation workflows to optimize segmentation performance and support reliable medical image analysis.
-
2021 - 2022 Tainan, Taiwan
Process Engineer
Taiwan Semiconductor Manufacturing Co., Ltd. (TSMC)
- Supported semiconductor manufacturing operations by monitoring equipment performance, identifying process issues, and contributing to production reliability and efficiency.
- Contributed to process improvement efforts to enhance operational stability and system reliability in a high-performance production environment.
- Conducted risk assessments and collaborated with cross-functional teams to resolve equipment- and process-related issues while supporting quality and safety standards.
-
2019 - 2021 Taipei, Taiwan
Software Engineer Intern
Test Research Inc.
X-Ray Image Super-Resolution
- Developed and optimized a deep learning model for X-ray image super-resolution using Python, TensorFlow, and TensorLayer.
- Improved image quality and enlargement efficiency through model design, experimentation, and performance tuning.
- Benchmarked the system against five classical AI models and four interpolation methods, demonstrating competitive quality and runtime performance.
-
2018 - 2018 Taichung, Taiwan
Software Engineer Intern
Chung Shan Medical University Hospital
Automated Head and Neck CT Organ Contouring
- Developed a Python-based automated organ contouring system for head and neck CT images.
- Built an end-to-end workflow for CT image conversion, organ segmentation, and RS file generation to support radiation oncology applications.
- Applied GAN-based deep learning to automate organ delineation and improve workflow consistency.
- Reduced organ selection and contouring time by 70%, improving efficiency for clinical staff.
-
2016 - 2018 Taichung, Taiwan
Developer
Chung Shan Medical University / Chung Shan Medical University Hospital
Document and Workflow Management Systems
- Designed and developed database-driven web platforms using PHP, JavaScript, HTML, CSS, MySQL, and OpenSSL to support document management, review workflows, admissions outreach, and administrative processes.
- Built role-specific workflow and automation features that improved process visibility, reduced manual workload, and supported institution-specific operational requirements.
- Automated internship placement and academic document workflows, reducing selected processing tasks from approximately one week to under one minute.
Publications
-
2026 ReDiPrune: Relevance-Diversity Pre-Projection Token Pruning for Efficient Multimodal LLMs
arXiv preprint
Developed a training-free token pruning framework for multimodal LLMs that improves inference efficiency by pruning visual tokens before the vision-language projector while preserving query-relevant and non-redundant information.
-
2026 RatSeizure: A Benchmark and Saliency-Context Transformer for Rat Seizure Localization
arXiv preprint
Built a benchmark and deep learning framework for rat seizure localization from video, supporting scalable and reproducible behavioral analysis through computer vision and temporal modeling.
-
2025 IntelliCardiac: An Intelligent Platform for Cardiac Image Segmentation and Classification
In 2025 IEEE 8th International Conference on Multimedia Information Processing and Retrieval (MIPR)
Built a full-stack AI platform for cardiac image segmentation, disease classification, and real-time visualization, enabling end-to-end clinical decision-support workflows.
-
2025 Spotting the Fakes: A Deep Dive into GAN-Generated Face Detection
ACM Transactions on Multimedia Computing, Communications and Applications
Surveyed GAN-generated face detection methods, datasets, evaluation settings, and open challenges across deep learning, physics-based, and physiology-based approaches.
-
2024 UU-Mamba: Uncertainty-aware U-Mamba for Cardiac Image Segmentation
In 2024 IEEE 7th International Conference on Multimedia Information Processing and Retrieval (MIPR)
Developed an uncertainty-aware cardiac image segmentation framework that improves robustness, generalization, and segmentation accuracy through Sharpness-Aware Minimization and uncertainty-guided loss design.
Awards
-
2018 Artificial Intelligence and Robotics Creative Design Competition
AI and Robotics Creative Design Competition
Third Prize
-
2018 Creative Dredging Engineering & Monitoring Technology Competition
Creative Dredging Engineering & Monitoring Technology Competition
Honorable Mention
-
2018 National Creativity Competition of Electronic Design (14th)
National Creativity Competition of Electronic Design
Honorable Mention
-
2018 Creative Software Application Content Award
Creative Software Application Competition
Honorable Mention
-
2017 Taipei Medical University Venture Capital Investment Contest
Taipei Medical University
First Prize and Best Popularity Award
-
2017 International Conference on Medical and Health Informatics
Incidence and Clinical Features of Multiple Primary Malignant Neoplasms: A 5-Year Retrospective Analysis
First Prize
-
2013 Korea International Women's Invention Exposition (KIWIE)
KIWIE
Gold Prize
-
2011 Korea International Women's Invention Exposition (KIWIE)
KIWIE
German Special Prize