Model Privacy

Model privacy in machine learning focuses on protecting sensitive information embedded within or accessible through trained models, aiming to prevent unauthorized data extraction or model replication. Current research emphasizes mitigating vulnerabilities in various architectures, including large language models (LLMs) and convolutional neural networks (CNNs), through techniques like differential privacy, homomorphic encryption, and zero-knowledge proofs, often addressing specific attack vectors such as membership inference and model stealing. These efforts are crucial for responsible AI deployment, balancing the benefits of powerful models with the need to safeguard user data and intellectual property.

Papers

August 1, 2024

Pathway to Secure and Trustworthy 6G for LLMs: Attacks, Defense, and Opportunities
Sunder Ali Khowaja, Parus Khuwaja, Kapal Dev, Hussam Al Hamadi, Engin Zeydan
Large Language Model Membership Inference Attack New Attack Biological Pathway Model Privacy

July 2, 2024

Footprints of Data in a Classifier Model: The Privacy Issues and Their Mitigation through Data Obfuscation
Payel Sadhukhan, Tanujit Chakraborty
Raw Data Footprint Extraction Privacy Issue Model Privacy Protected Attribute Data Obfuscation AI Vulnerability

June 16, 2024

Promoting Data and Model Privacy in Federated Learning through Quantized LoRA
JianHao Zhu, Changze Lv, Xiaohua Wang, Muling Wu, Wenhao Liu, Tianlong Li, Zixuan Ling, Cenyuan Zhang, Xiaoqing Zheng, Xuanjing Huang
Large Language Model Federated Learning Raw Data Gradient Estimation Model Privacy Quantization Loss

March 17, 2024

Pencil: Private and Extensible Collaborative Learning without the Non-Colluding Assumption
Xuanqi Liu, Zhuotao Liu, Qi Li, Ke Xu, Mingwei Xu
Homomorphic Encryption Collaborative Learning Privacy Preserving Machine Learning Private Model Training Model Privacy Secure Multiparty Computation Tacit Collusion Pen Like Object

February 9, 2024

Trust the Process: Zero-Knowledge Machine Learning to Enhance Trust in Generative AI Interactions
Bianca-Mihaela Ganescu, Jonathan Passerat-Palmbach
Complex Process Building Trust Zero Knowledge Model Privacy Conversational Generative

January 21, 2024

Tempo: Confidentiality Preservation in Cloud-Based Neural Network Training
Rongwu Xu, Zhixuan Fang
Deep Learning Learning Platform Cloud Based Cost Model Model Privacy Tempo Estimation

December 18, 2023

Model Stealing Attack against Recommender System
Zhihao Zhu, Rui Fan, Chenwang Wu, Yi Yang, Defu Lian, Enhong Chen
Adversarial Attack Recommender System Privacy Attack Model Privacy Model Stealing Attack

November 12, 2023

Tunable Soft Prompts are Messengers in Federated Learning
Chenhe Dong, Yuexiang Xie, Bolin Ding, Ying Shen, Yaliang Li
Federated Setting Soft Prompt Model Privacy Federated Learning Training

November 5, 2023

Differentially Private Pre-Trained Model Fusion using Decentralized Federated Graph Matching
Qian Chen, Yiqiang Chen, Xinlong Jiang, Teng Zhang, Weiwei Dai, Wuliang Huang, Zhen Yan, Bo Ye
Local Differential Privacy Graph Matching Model Fusion Model Privacy Model Exchange

October 22, 2023

MoPe: Model Perturbation-based Privacy Attacks on Language Models
Marvin Li, Jason Wang, Jeffrey Wang, Seth Neel
Language Model Model Privacy

October 10, 2023

Be Careful What You Smooth For: Label Smoothing Can Be a Privacy Shield but Also a Catalyst for Model Inversion Attacks
Lukas Struppek, Dominik Hintersdorf, Kristian Kersting
Label Smoothing Regularization Technique Model Inversion Attack Model Privacy Label Only Model Inversion Attack

September 22, 2023

Privacy Assessment on Reconstructed Images: Are Existing Evaluation Metrics Faithful to Human Perception?
Xiaoxiao Sun, Nidham Gazagnadou, Vivek Sharma, Lingjuan Lyu, Hongdong Li, Liang Zheng
Evaluation Metric PRIvacy Leakage Human Perception Model Privacy Reconstructed Image Image Quality Metric Privacy Evaluation

July 21, 2023

Epsilon*: Privacy Metric for Machine Learning Models
Diana M. Negoescu, Humberto Gonzalez, Saad Eddin Al Orjany, Jilei Yang, Yuliia Lut, Rahul Tandra, Xiaowen Zhang, Xinyi Zheng, Zach Douglas, Vidita Nolkha, Parvez Ahammad, Gennady Samorodnitsky
Machine Learning Model Differential Privacy Privacy Risk Empirical Privacy Model Privacy

June 1, 2023

TMI! Finetuned Models Leak Private Information from their Pretraining Data
John Abascal, Stanley Wu, Alina Oprea, Jonathan Ullman
Machine Learning Transfer Learning Differential Privacy Pre Training Membership Inference Attack Model Privacy

May 24, 2023

Privacy Implications of Retrieval-Based Language Models
Yangsibo Huang, Samyak Gupta, Zexuan Zhong, Kai Li, Danqi Chen
Language Model Private Data Model Privacy

April 26, 2023

FedVS: Straggler-Resilient and Privacy-Preserving Vertical Federated Learning for Split Models
Songze Li, Duanyi Yao, Jin Liu
Vertical Federated Learning Model Privacy Secret Sharing Model Splitting VFL Algorithm Information Theoretic Privacy

April 11, 2023

Echo of Neighbors: Privacy Amplification for Personalized Private Federated Learning with Shuffle Model
Yixuan Liu, Suyun Zhao, Li Xiong, Yuhan Liu, Hong Chen
Personalized Federated Learning Thy Neighbor Shuffle Model Privacy Amplification Model Privacy

March 7, 2023

Bootstrap The Original Latent: Learning a Private Model from a Black-box Model
Shuai Wang, Daoan Zhang, Jianguo Zhang, Weiwei Zhang, Rui Li
Black Box Model Intermediate Latent Private Model Model Privacy

October 20, 2022

How Does a Deep Learning Model Architecture Impact Its Privacy? A Comprehensive Study of Privacy Attacks on CNNs and Transformers
Guangsheng Zhang, Bo Liu, Huan Tian, Tianqing Zhu, Ming Ding, Wanlei Zhou
Convolutional Neural Network Deep Learning Model Transformer Megatron Decepticons Privacy Policy Neural Network Architecture Privacy Risk Privacy Attack Deep Learning Technology Model Privacy

July 21, 2022

Careful What You Wish For: on the Extraction of Adversarially Trained Models
Kacem Khaled, Gabriela Nicolescu, Felipe Gohring de Magalhães
Adversarial Example Full Model Adversarial Training Process Extraction Model Privacy Extraction Attack Adversarial Capability

Model Privacy

Papers

Pathway to Secure and Trustworthy 6G for LLMs: Attacks, Defense, and Opportunities

Footprints of Data in a Classifier Model: The Privacy Issues and Their Mitigation through Data Obfuscation

Promoting Data and Model Privacy in Federated Learning through Quantized LoRA

Pencil: Private and Extensible Collaborative Learning without the Non-Colluding Assumption

Trust the Process: Zero-Knowledge Machine Learning to Enhance Trust in Generative AI Interactions

Tempo: Confidentiality Preservation in Cloud-Based Neural Network Training

Model Stealing Attack against Recommender System

Tunable Soft Prompts are Messengers in Federated Learning

Differentially Private Pre-Trained Model Fusion using Decentralized Federated Graph Matching

MoPe: Model Perturbation-based Privacy Attacks on Language Models

Be Careful What You Smooth For: Label Smoothing Can Be a Privacy Shield but Also a Catalyst for Model Inversion Attacks

Privacy Assessment on Reconstructed Images: Are Existing Evaluation Metrics Faithful to Human Perception?

Epsilon*: Privacy Metric for Machine Learning Models

TMI! Finetuned Models Leak Private Information from their Pretraining Data

Privacy Implications of Retrieval-Based Language Models

FedVS: Straggler-Resilient and Privacy-Preserving Vertical Federated Learning for Split Models

Echo of Neighbors: Privacy Amplification for Personalized Private Federated Learning with Shuffle Model

Bootstrap The Original Latent: Learning a Private Model from a Black-box Model

How Does a Deep Learning Model Architecture Impact Its Privacy? A Comprehensive Study of Privacy Attacks on CNNs and Transformers

Careful What You Wish For: on the Extraction of Adversarially Trained Models