Privacy Ripple Effects from Adding or Removing Personal Information in Language Model Training
From Google TechTalks · 48:47
Show by Google TechTalks
A Google TechTalk, 2025-05-14, presented by Jaydeep BorkarPrivacy in ML Seminar. ABSTRACT: Due to the sensitive nature of personally identifiable information (PII), its owners may have the authority to control its inclusion or request its removal from large-language model (LLM) training. Beyond this, PII may be added or removed from training datasets due to evolving dataset curation techniques, because they were newly scraped for retraining, or because they were included in a new downstream fine-tuning stage. We find that the amount and ease of PII memorization is a dynamic property of a model that evolves throughout training pipelines and depends on commonly altered design choices. We characterize three such novel phenomena: (1) similar-appearing PII seen later in training can elicit memorization of earlier-seen sequences in what we call assisted memorization, and this is a significant factor (in our settings, up to 1/3); (2) adding PII can increase memorization of other PII significantly (in our settings, as much as 7.5x); and (3) removing PII can lead to other PII being memorized. Model creators should consider these first- and second-order privacy risks when training models to avoid the risk of new PII regurgitation.Uploaded 26-01-26
More episodes of Google TechTalks
More episodes
- DSD: A Hybrid Analysis Tool for Bug Finding
- Chasing the Constants and its Implications in Differential Privacy
- Making Great Decisions
- Streaming Private Continual Counting via Binning
- Distributed Testing with SmartFrog
- Differentially Private Synthetic Data without Training
- A Dynamic Network Layer for Advanced Cloud Computing
- Data Representation/Laplace Operator
- Persistent Pre-Training Poisoning of LLMs
- Threat Models for Memorization: Privacy, Copyright, and Everything In-Between
PodSearch is a free, independent podcast search engine. Read about PodSearch, the privacy policy or contact us.