Zhangheng Li

ORCID: 0000-0002-1837-484X
Publications
Citations
Views
---
Saved
---
About
Contact & Profiles
Research Areas
  • Concrete and Cement Materials Research
  • Magnesium Oxide Properties and Applications
  • Privacy-Preserving Technologies in Data
  • Innovative concrete reinforcement materials
  • Stochastic Gradient Optimization Techniques
  • Metallurgical Processes and Thermodynamics
  • Recycling and utilization of industrial and municipal waste in materials production
  • Concrete Corrosion and Durability
  • Usability and User Interface Design
  • Advanced Photocatalysis Techniques
  • Layered Double Hydroxides Synthesis and Applications
  • Generative Adversarial Networks and Image Synthesis
  • Recycled Aggregate Concrete Performance
  • Concrete Properties and Behavior

The University of Texas at Austin
2024

Southern Marine Science and Engineering Guangdong Laboratory (Zhuhai)
2022-2024

Southern Marine Science and Engineering Guangdong Laboratory (Guangzhou)
2022-2024

Sun Yat-sen University
2022-2024

While diffusion models have recently demonstrated remarkable progress in generating realistic images, privacy risks also arise: published or APIs could generate training images and thus leak privacy-sensitive information. In this paper, we reveal a new risk, Shake-to-Leak (S2L), that fine-tuning the pre-trained with manipulated data can amplify existing risks. We demonstrate S2L occur various standard strategies for models, including concept-injection methods (DreamBooth Textual Inversion)...

10.1109/satml59370.2024.00010 article EN 2024-04-09

While diffusion models have recently demonstrated remarkable progress in generating realistic images, privacy risks also arise: published or APIs could generate training images and thus leak privacy-sensitive information. In this paper, we reveal a new risk, Shake-to-Leak (S2L), that fine-tuning the pre-trained with manipulated data can amplify existing risks. We demonstrate S2L occur various standard strategies for models, including concept-injection methods (DreamBooth Textual Inversion)...

10.48550/arxiv.2403.09450 preprint EN arXiv (Cornell University) 2024-03-14

Building a generalist model for user interface (UI) understanding is challenging due to various foundational issues, such as platform diversity, resolution variation, and data limitation. In this paper, we introduce Ferret-UI 2, multimodal large language (MLLM) designed universal UI across wide range of platforms, including iPhone, Android, iPad, Webpage, AppleTV. on the foundation Ferret-UI, 2 introduces three key innovations: support multiple types, high-resolution perception through...

10.48550/arxiv.2410.18967 preprint EN arXiv (Cornell University) 2024-10-24

This paper provided a new alkali activator for steel slag cementitious material - layered sodium disilicate (LSD), and focused on its dissolution-precipitation mechanism based ion exchange capacity of LSD. The results showed that LSD could act as an effective slag, the higher compressive strength pastes (S3(14.9MPa) S10(22.3MPa) increased by 222% 332% respectively) were achieved matrix with much denser pore structure was formed. solution hydration heat analysis indicated play role Ca2+...

10.2139/ssrn.4607981 preprint EN 2023-01-01
Coming Soon ...