Conference

We have 3 papers accepted by NeurIPS'2025 on privacy-preserving AI (PPAI) and efficient embodied AI. The 2 papers on PPAI are “CryptoMoE: Privacy-Preserving and Scalable Mixture of Experts Inference via Balanced Expert Routing” and “MPCache: MPC-Friendly KV Cache Eviction for Efficient Private LLM Inference”. The paper on embodied AI is “EfficientNav: Towards On-Device Object-Goal Navigation with Navigation Map Caching and Retrieval”.

Our latest survey paper on privacy-preserving ML (PPML) is posted on Arxiv. The paper summarizes the recent progress on efficient PPML across algorithm, protocol, and system levels. The title of our paper is “Towards Efficient Privacy-Preserving Machine Learning: A Systematic Review from Protocol, Model, and System Perspectives”. We also prepare a Github repo to continue tracking all the relevant papers in the field “https://github.com/PKU-SEC-Lab/Awesome-PPML-Papers”.

Congratulations to Zizhuo Fu (junior student in the SEC Lab) for winning the prestigious SenseTime Scholarship. Zizhuo joined the SEC Lab since his sophomore year and will continue to pursue his PhD degree under the joint supervision of Prof. Runsheng Wang and me.

We have two papers on hardware acceleration for privacy-preserving AI accepted, including “Ironman: Accelerating Oblivious Transfer Extension for Privacy-Preserving AI with Near-Memory Processing” accepted by MICRO'2025 and “Matrix: Multi-Cipher Structures Dataflow for Parallel and Pipelined TFHE Accelerator” (led by Prof. Ling Liang) accepted by TACO'2025.

Five papers on efficient AI and one special session proposal on privacy-preserving AI are accepted by ICCAD'2025.

One paper on privacy-preserving Transformer inference is accepted by Usenix Security'2025 as a regular paper. The title of the paper is “Breaking the Layer Barrier: Remodeling Private Transformer Inference with Hybrid CKKS and MPC”.

One collaboration paper on “Swift: Fast Secure Neural Network Inference with Fully Homomorphic Encryption” is accepted by IEEE TIFS'2025.

Four papers on efficient LLM are accepted by DAC'2024 as regular papers.

Four papers, including 3 on efficient AI and 1 on privacy-preserving AI, are accepted by DATE'2025 as regular papers.

Two papers on “PrivCirNet: Efficient Private Inference via Block Circulant Transformation” and “ArkVale: Efficient Generative LLM Inference with Recallable Key-Value Eviction” are accpted by NeurIPs'2024.