8 papers
Sequential Data Poisoning in LLM Post-Training
Jack Sanderson, Yihan Wang, Xiaoqian Lu +2
LLM post-training proceeds through multiple stages, e.g., supervised fine-tuning (SFT) followed by reinforcement learning from human feedback (RLHF) or direct preference optimizati…
DriftXpress: Faster Drifting Models via Projected RKHS Fields
Ali Falahati, Elliot Creager, Gautam Kamath +1
Drifting Models have emerged as a new paradigm for one-step generative modeling, achieving strong image quality without iterative inference. The premise is to replace the iterative…
Are Targeted Data Poisoning Attacks as Effective as We Think?
William Xu, Chenyu Zhang, Yihan Wang +5
Targeted data poisoning attacks manipulate model predictions on specific test samples by injecting malicious data into training. Yet existing evaluations report average attack succ…
Demystifying Foreground-Background Memorization in Diffusion Models
Jimmy Z. Di, Yiwei Lu, Yaoliang Yu +3
Diffusion models (DMs) memorize training images and can reproduce near-duplicates during generation. Current detection methods identify verbatim memorization but fail to capture tw…
BridgePure: Limited Protection Leakage Can Break Black-Box Data Protection
Yihan Wang, Yiwei Lu, Xiao-Shan Gao +2
Availability attacks, or unlearnable examples, are defensive techniques that allow data owners to modify their datasets in ways that prevent unauthorized machine learning models fr…
Machine Unlearning Fails to Remove Data Poisoning Attacks
Martin Pawelczyk, Jimmy Z. Di, Yiwei Lu +3
We revisit the efficacy of several practical methods for approximate machine unlearning developed for large-scale deep learning. In addition to complying with data deletion request…
Disguised Copyright Infringement of Latent Diffusion Models
Yiwei Lu, Matthew Y. R. Yang, Zuoqiu Liu +2
Copyright infringement may occur when a generative model produces samples substantially similar to some copyrighted data that it had access to during the training phase. The notion…
Indiscriminate Data Poisoning Attacks on Pre-trained Feature Extractors
Yiwei Lu, Matthew Y. R. Yang, Gautam Kamath +1
Machine learning models have achieved great success in supervised learning tasks for end-to-end training, which requires a large amount of labeled data that is not always feasible.…