Research Highlights

Privacy-Preserving Large Models

In an era where artificial intelligence shapes our world, ensuring the security of large language models like ChatGPT, Llama3, and Qwen is paramount. Our advanced evaluation algorithms empower us to uncover vulnerabilities with 10 times greater efficiency, leaving no stone unturned in the quest for privacy.

We leverage a powerful combination of cutting-edge techniques, including RedTeaming, Fuzzing, and Jailbreaking, to thoroughly identify potential privacy leaks. Our rigorous approach guarantees that we are at the forefront of AI security.

But we don’t just identify vulnerabilities—we excel in defending against them. With innovative solutions like differential privacy and unlearning, we provide robust protection that outperforms existing methods, ensuring that your AI applications are not only powerful but also secure.

Image 1 Image 2 Image 3

Split Learning Solutions

We are proud to introduce our groundbreaking open-source solutions for split learning, designed to elevate privacy across various applications, including diffusion models, large language models, and face recognition systems. At the heart of our innovations lies Patch Shuffling, an advanced algorithm that sets a new standard in privacy guarantees.

With our cutting-edge method, you can trust that your data is protected at a level comparable to cryptographic standards, ensuring confidentiality without compromising on performance. Our solution delivers efficiency and accuracy that rivals traditional plain-text processing, enabling seamless integration into your AI projects.

Embrace the future of secure AI with our trusted solutions, where innovation meets unmatched privacy.

Image 1

Attacker cannot succesfully reconstruct users faces when our defense is presented.

Xu, Hengyuan, Liyao Xiang, Hangyu Ye, Dixi Yao, Pengzhi Chu, and Baochun Li. “Permutation Equivariance of Transformers and Its Applications.” In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pp. 5987-5996. 2024.

Federated Learning Solutions

We empower businesses to harness the power of machine learning without compromising privacy. Our cutting-edge federated learning platform allows customers to build and customize their own AI models—all without sharing sensitive data. By leveraging our efficient, user-friendly software solutions, you can innovate faster, streamline processes, and protect your personal information with complete confidence. Revolutionize your workflow and let our platform take your business to the next level.

On our platform, the memory usage can be reduced to only 1/3 of original consumption.