Cover Image

The Curse of Recursion: Training on Generated Data Makes Models Forget

Fuente: arXiv

Saved in:

Bibliographic Details
Main Authors:	Shumailov, Ilia, Shumaylov, Zakhar, Zhao, Yiren, Gal, Yarin, Papernot, Nicolas, Anderson, Ross
Format:	Preprint
Published:	2023
Subjects:	Machine Learning Artificial Intelligence Computation and Language Cryptography and Security Computer Vision and Pattern Recognition
Online Access:	Acceder al recurso
Tags:	Add Tag No Tags, Be the first to tag this record!

Similar Items

Architectural Neural Backdoors from First Principles
by: Langford, Harry, et al.
Published: (2024)

When Vision Fails: Text Attacks Against ViT and OCR
by: Boucher, Nicholas, et al.
Published: (2023)

Breach By A Thousand Leaks: Unsafe Information Leakage in `Safe' AI Responses
by: Glukhov, David, et al.
Published: (2024)

Generating Synthetic Data with Formal Privacy Guarantees: State of the Art and the Road Ahead
by: Schlegel, Viktor, et al.
Published: (2025)

Reasoning Introduces New Poisoning Attacks Yet Makes Them More Complicated
by: Foerster, Hanna, et al.
Published: (2025)

Watermarking Needs Input Repetition Masking
by: Khachaturov, David, et al.
Published: (2025)

UnUnlearning: Unlearning is not sufficient for content regulation in advanced generative AI
by: Shumailov, Ilia, et al.
Published: (2024)

ImpNet: Imperceptible and blackbox-undetectable backdoors in compiled neural networks
by: Clifford, Eleanor, et al.
Published: (2022)

Iteratively Prompting Multimodal LLMs to Reproduce Natural and AI-Generated Images
by: Naseh, Ali, et al.
Published: (2024)

BSPA: Exploring Black-box Stealthy Prompt Attacks against Image Generators
by: Tian, Yu, et al.
Published: (2024)

IAG: Input-aware Backdoor Attack on VLM-based Visual Grounding
by: Li, Junxian, et al.
Published: (2025)

Beyond Laplace and Gaussian: Exploring the Generalized Gaussian Mechanism for Private Machine Learning
by: Rinberg, Roy, et al.
Published: (2025)

Stealing User Prompts from Mixture of Experts
by: Yona, Itay, et al.
Published: (2024)

Beyond Labeling Oracles: What does it mean to steal ML models?
by: Shafran, Avital, et al.
Published: (2023)

VLMs Can Aggregate Scattered Training Patches
by: Zhou, Zhanhui, et al.
Published: (2025)

LLM Dataset Inference: Did you train on my dataset?
by: Maini, Pratyush, et al.
Published: (2024)

Leak and Learn: An Attacker's Cookbook to Train Using Leaked Data from Federated Learning
by: Zhao, Joshua C., et al.
Published: (2024)

Are GUI Agents Focused Enough? Automated Distraction via Semantic-level UI Element Injection
by: Yang, Wenkui, et al.
Published: (2026)

Image-Based Geolocation Using Large Vision-Language Models
by: Liu, Yi, et al.
Published: (2024)

MLLM-Protector: Ensuring MLLM's Safety without Hurting Performance
by: Pi, Renjie, et al.
Published: (2024)

Privacy-Preserving Federated Learning with Verifiable Fairness Guarantees
by: Ali, Mohammed Himayath, et al.
Published: (2026)

Visual Contextual Attack: Jailbreaking MLLMs with Image-Driven Context Injection
by: Miao, Ziqi, et al.
Published: (2025)

SlowBA: An efficiency backdoor attack towards VLM-based GUI agents
by: Li, Junxian, et al.
Published: (2026)

Rethinking Bottlenecks in Safety Fine-Tuning of Vision Language Models
by: Ding, Yi, et al.
Published: (2025)

Contextual Image Attack: How Visual Context Exposes Multimodal Safety Vulnerabilities
by: Xiong, Yuan, et al.
Published: (2025)

Effective and Efficient Adversarial Detection for Vision-Language Models via A Single Vector
by: Huang, Youcheng, et al.
Published: (2024)

Are Vision-Language Models Safe in the Wild? A Meme-Based Benchmark Study
by: Lee, DongGeon, et al.
Published: (2025)

Self-adaptive Dataset Construction for Real-World Multimodal Safety Scenarios
by: Qu, Jingen, et al.
Published: (2025)

Doubly-Universal Adversarial Perturbations: Deceiving Vision-Language Models Across Both Images and Text with a Single Perturbation
by: Kim, Hee-Seon, et al.
Published: (2024)

Evolving Contextual Safety in Multi-Modal Large Language Models via Inference-Time Self-Reflective Memory
by: Zhang, Ce, et al.
Published: (2026)

Forget-MI: Machine Unlearning for Forgetting Multimodal Information in Healthcare Settings
by: Hardan, Shahad, et al.
Published: (2025)

Inexact Unlearning Needs More Careful Evaluations to Avoid a False Sense of Privacy
by: Hayes, Jamie, et al.
Published: (2024)

Interpreting the Repeated Token Phenomenon in Large Language Models
by: Yona, Itay, et al.
Published: (2025)

Rethinking Machine Unlearning in Image Generation Models
by: Liu, Renyang, et al.
Published: (2025)

Recovering the Pre-Fine-Tuning Weights of Generative Models
by: Horwitz, Eliahu, et al.
Published: (2024)

Reconstructing Training Data From Real World Models Trained with Transfer Learning
by: Oz, Yakir, et al.
Published: (2024)

SafeGen: Mitigating Sexually Explicit Content Generation in Text-to-Image Models
by: Li, Xinfeng, et al.
Published: (2024)

One Pic is All it Takes: Poisoning Visual Document Retrieval Augmented Generation with a Single Image
by: Shereen, Ezzeldin, et al.
Published: (2025)

When Machine Unlearning Meets Retrieval-Augmented Generation (RAG): Keep Secret or Forget Knowledge?
by: Wang, Shang, et al.
Published: (2024)

Color Matters: Demosaicing-Guided Color Correlation Training for Generalizable AI-Generated Image Detection
by: Zhong, Nan, et al.
Published: (2026)