Image de couverture de livre

TMI! Finetuned Models Leak Private Information from their Pretraining Data

Fuente: arXiv

Enregistré dans:

Détails bibliographiques
Auteurs principaux:	Abascal, John, Wu, Stanley, Oprea, Alina, Ullman, Jonathan
Format:	Preprint
Publié:	2023
Sujets:	Machine Learning Cryptography and Security
Accès en ligne:	Acceder al recurso
Tags:	Ajouter un tag Pas de tags, Soyez le premier à ajouter un tag!

Documents similaires

Black-Box Privacy Attacks on Shared Representations in Multitask Learning
par: Abascal, John, et autres
Publié: (2025)

Reconstruction of Personally Identifiable Information from Supervised Finetuned Models
par: Furukawa, Sae, et autres
Publié: (2026)

Lower Bounds for Public-Private Learning under Distribution Shift
par: Setlur, Amrith, et autres
Publié: (2025)

UTrace: Poisoning Forensics for Private Collaborative Learning
par: Rose, Evan, et autres
Publié: (2024)

How to Make the Gradients Small Privately: Improved Rates for Differentially Private Non-Convex Optimization
par: Lowy, Andrew, et autres
Publié: (2024)

Adversarial Inception Backdoor Attacks against Reinforcement Learning
par: Rathbun, Ethan, et autres
Publié: (2024)

SleeperNets: Universal Backdoor Poisoning Attacks Against Reinforcement Learning Agents
par: Rathbun, Ethan, et autres
Publié: (2024)

Smooth Lower Bounds for Differentially Private Algorithms via Padding-and-Permuting Fingerprinting Codes
par: Peter, Naty, et autres
Publié: (2023)

Cascading Adversarial Bias from Injection to Distillation in Language Models
par: Chaudhari, Harsh, et autres
Publié: (2025)

The Sample Complexity of Membership Inference and Privacy Auditing
par: Haghifam, Mahdi, et autres
Publié: (2025)

Exploiting Leaderboards for Large-Scale Distribution of Malicious Models
par: Suri, Anshuman, et autres
Publié: (2025)

Text-to-Image Models Leave Identifiable Signatures: Implications for Leaderboard Security
par: Naseh, Ali, et autres
Publié: (2025)

Syntax- and Compilation-Preserving Evasion of LLM Vulnerability Detectors
par: Sun, Luze, et autres
Publié: (2026)

Model-agnostic clean-label backdoor mitigation in cybersecurity environments
par: Severi, Giorgio, et autres
Publié: (2024)

Privacy in Metalearning and Multitask Learning: Modeling and Separations
par: Aliakbarpour, Maryam, et autres
Publié: (2024)

Backdoor Attacks in Peer-to-Peer Federated Learning
par: Syros, Georgios, et autres
Publié: (2023)

Beware Untrusted Simulators -- Reward-Free Backdoor Attacks in Reinforcement Learning
par: Rathbun, Ethan, et autres
Publié: (2026)

Phantom: General Backdoor Attacks on Retrieval Augmented Language Generation
par: Chaudhari, Harsh, et autres
Publié: (2024)

Quantamination: Dynamic Quantization Leaks Your Data Across the Batch
par: Foerster, Hanna, et autres
Publié: (2026)

Quantitative Resilience Modeling for Autonomous Cyber Defense
par: Cadet, Xavier, et autres
Publié: (2025)

R1dacted: Investigating Local Censorship in DeepSeek's R1 Language Model
par: Naseh, Ali, et autres
Publié: (2025)

Position: Considerations for Differentially Private Learning with Large-Scale Public Pretraining
par: Tramèr, Florian, et autres
Publié: (2022)

Toward Principled LLM Safety Testing: Solving the Jailbreak Oracle Problem
par: Lin, Shuyi, et autres
Publié: (2025)

Attacks and Mitigations for Distributed Governance of Agentic AI under Byzantine Adversaries
par: Laws, Matthew D., et autres
Publié: (2026)

Thought-Transfer: Indirect Targeted Poisoning Attacks on Chain-of-Thought Reasoning Models
par: Chaudhari, Harsh, et autres
Publié: (2026)

ACE: A Security Architecture for LLM-Integrated App Systems
par: Li, Evan, et autres
Publié: (2025)

Differentially Private Federated $k$-Means Clustering with Server-Side Data
par: Scott, Jonathan, et autres
Publié: (2025)

Privacy Backdoors: Stealing Data with Corrupted Pretrained Models
par: Feng, Shanglun, et autres
Publié: (2024)

One-shot Empirical Privacy Estimation for Federated Learning
par: Andrew, Galen, et autres
Publié: (2023)

DeepLeak: Privacy Enhancing Hardening of Model Explanations Against Membership Leakage
par: Hmida, Firas Ben, et autres
Publié: (2026)

Oracle-Efficient Differentially Private Learning with Public Data
par: Block, Adam, et autres
Publié: (2024)

From Data Leak to Secret Misses: The Impact of Data Leakage on Secret Detection Models
par: Soltaniani, Farnaz, et autres
Publié: (2026)

Shake to Leak: Fine-tuning Diffusion Models Can Amplify the Generative Privacy Risk
par: Li, Zhangheng, et autres
Publié: (2024)

User Inference Attacks on Large Language Models
par: Kandpal, Nikhil, et autres
Publié: (2023)

Pre-training Differentially Private Models with Limited Public Data
par: Bu, Zhiqi, et autres
Publié: (2024)

CoinPress: Practical Private Mean and Covariance Estimation
par: Biswas, Sourav, et autres
Publié: (2020)

Bounding the Excess Risk for Linear Models Trained on Marginal-Preserving, Differentially-Private, Synthetic Data
par: Zhou, Yvonne, et autres
Publié: (2024)

PoolFlip: A Multi-Agent Reinforcement Learning Security Environment for Cyber Defense
par: Cadet, Xavier, et autres
Publié: (2025)

SAGA: A Security Architecture for Governing AI Agentic Systems
par: Syros, Georgios, et autres
Publié: (2025)

Private Mean Estimation with Person-Level Differential Privacy
par: Agarwal, Sushant, et autres
Publié: (2024)