Controllable and Efficient Generative Models
- Zero-shot Video Editing
- Video Diffusion / Video DiT Models
- Efficient Diffusion / DiT Inference
- Model Compression and Quantization
- Structure-preserving Generation
About
I am a direct Ph.D. student in Electronics Science and Technology at Zhejiang University, advised by Prof. Kejie Huang.
My research lies at the intersection of generative modeling and efficient AI systems. I study controllable video generation and editing for diffusion/Video DiT models, as well as efficiency techniques such as model compression and inference acceleration.
Before focusing on AI algorithms, I was trained in Microelectronics Science and Engineering. This background allows me to approach generative models from a system perspective, bridging algorithm design with efficient hardware-aware deployment.
Research Interests
Algorithm-Hardware Co-design
- Hardware-aware Generative Models
- Efficient Deployment of Foundation Models
- System-Algorithm Co-design for Generative AI
Multimodal and Unified Generative Models
- Multimodal Large Language Models (MLLMs)
- Unified Models for Text, Image, and Video Generation
- Cross-modal Reasoning and Generation
Selected Publications
FreeMask: Rethinking the Importance of Attention Masks for Zero-Shot Video Editing
AAAI 2025
DFVEdit: Conditional Delta Flow Vector for Zero-shot Video Editing
Submitted (NeurIPS 2025, OpenReview)
A Near-array Convolution Computing Scheme Based on WSe2
PIERS 2021 · Best Student Paper Award (Optics and Photonics Section, 5th)
DeQuan: Decoupled Quantization for Efficient DETRs
Manuscript under review
Only the Pivotal Data Matters: An Acceleration Method for Quantization Aware Training
Manuscript under review
Research Experience
Video Editing (Jul 2023 - Present)
Developed zero-shot video editing methods for text-to-video diffusion and Video DiT models, with focus on structure preservation, controllability, and efficiency.
Model Compression and Inference Acceleration (Jul 2022 - Present)
Proposed dataset condensation and decoupled quantization methods to improve QAT efficiency and reduce quantization errors for Transformer detectors.
Low-power In-sensor Convolutional Computing (Mar 2021 - Sep 2021)
Explored hardware-software co-design with photodetector sensor arrays for low-power, low-latency near-sensor computing.
Education
Zhejiang University
Ph.D. in Electronics Science and Technology (Direct Ph.D. Program), 2021 - Present
Zhejiang University
B.Eng. in Microelectronics Science and Technology, 2017 - 2021 (strong academic performance)
Projects
Contact
Email: cailingling22@zju.edu.cn
Links: GitHub · OpenReview