Sequential LLM defense deployment leads to risk exacerbation in 38.9% of cases due to anti-aligned updates in shared critical layers, addressed by conflict-guided layer freezing.
Large language models can be strong differentially private learners
13 Pith papers cite this work. Polarity classification is still indexing.
13
Pith papers citing it
representative citing papers
A new probing framework detects moderate parametric memorization signals in tabular in-context learning models under single-task fine-tuning, strongest on low-cardinality tasks, but signals largely disappear under realistic training.
Balanced Iteration Subsampling achieves stronger privacy amplification than Poisson subsampling in DP-SGD by eliminating participation variance while keeping uniform marginal participation.
DPrivBench is a new benchmark for evaluating LLMs on differential privacy reasoning, with results showing good performance on textbook mechanisms but substantial failures on advanced algorithms.
DP-OPD achieves lower perplexity than DP fine-tuning and synthesis-based private distillation under ε=2.0 by enforcing DP-SGD solely on the student during on-policy training with a frozen teacher.
Introduces MM-Privacy dataset and evaluations showing MLLMs leak sensitive data from images in various tasks, highlighting task inconsistency effects.
CSLR aligns unordered private replay lists from clients using public anchor sentence signatures, yielding 3.9-5.6 point gains on continual NLP tasks at ε=4 over non-CSLR DP baselines.
DP-SGD-RC applies Hutchinson and Hutch++ estimators to approximate per-sample gradient norms for clipping in DP-SGD, claiming competitive privacy noise multipliers and utility on Llama 3.2-1B with reduced memory.
DP-SelFT improves the privacy-utility trade-off for LLM fine-tuning by selecting robust layer subsets via DP synthetic data and perturbation-matched evaluation.
GroupGPT decouples intervention timing from response generation via edge-cloud collaboration for multi-user chats, scoring 4.72/5 on the new MUIR benchmark of 2500 segments while cutting token use by up to 3x and adding privacy sanitization.
DP-GRAPE reduces memory in differentially private neural network training by using random Gaussian projections on gradients instead of SVD, achieving comparable privacy-utility tradeoffs to DP-SGD and scaling to 6.7B parameter models.
A data-centric survey finds that only information-flow control covers compositional and cross-session leakage in LLM agents and that no single benchmark tests an agent across all its data surfaces under one policy.
HERALD selectively encrypts sensitive tokens via medical NER, POS policies, and deterministic ciphertext substitution to enable privacy-preserving clinical LLM use while recovering near-plaintext task performance.
citing papers explorer
-
Defenses at Odds: Measuring and Explaining Defense Conflicts in Large Language Models
Sequential LLM defense deployment leads to risk exacerbation in 38.9% of cases due to anti-aligned updates in shared critical layers, addressed by conflict-guided layer freezing.
-
Probing Memorization of Tabular In-Context Learning
A new probing framework detects moderate parametric memorization signals in tabular in-context learning models under single-task fine-tuning, strongest on low-cardinality tasks, but signals largely disappear under realistic training.
-
Less Random, More Private: What is the Optimal Subsampling Scheme for DP-SGD?
Balanced Iteration Subsampling achieves stronger privacy amplification than Poisson subsampling in DP-SGD by eliminating participation variance while keeping uniform marginal participation.
-
DPrivBench: Benchmarking LLMs' Reasoning for Differential Privacy
DPrivBench is a new benchmark for evaluating LLMs on differential privacy reasoning, with results showing good performance on textbook mechanisms but substantial failures on advanced algorithms.
-
DP-OPD: Differentially Private On-Policy Distillation for Language Models
DP-OPD achieves lower perplexity than DP fine-tuning and synthesis-based private distillation under ε=2.0 by enforcing DP-SGD solely on the student during on-policy training with a frozen teacher.
-
Unveiling Privacy Risks in Multi-modal Large Language Models: Task-specific Vulnerabilities and Mitigation Challenges
Introduces MM-Privacy dataset and evaluations showing MLLMs leak sensitive data from images in various tasks, highlighting task inconsistency effects.
-
Canonicalized Stable-List Replay for Private Federated Continual Learning over Language-Model Embeddings
CSLR aligns unordered private replay lists from clients using public anchor sentence signatures, yielding 3.9-5.6 point gains on continual NLP tasks at ε=4 over non-CSLR DP baselines.
-
Efficient DP-SGD for LLMs with Randomized Clipping
DP-SGD-RC applies Hutchinson and Hutch++ estimators to approximate per-sample gradient norms for clipping in DP-SGD, claiming competitive privacy noise multipliers and utility on Llama 3.2-1B with reduced memory.
-
DP-SelFT: Differentially Private Selective Fine-Tuning for Large Language Models
DP-SelFT improves the privacy-utility trade-off for LLM fine-tuning by selecting robust layer subsets via DP synthetic data and perturbation-matched evaluation.
-
GroupGPT: A Token-efficient and Privacy-preserving Agentic Framework for Multi-User Chat Assistant
GroupGPT decouples intervention timing from response generation via edge-cloud collaboration for multi-user chats, scoring 4.72/5 on the new MUIR benchmark of 2500 segments while cutting token use by up to 3x and adding privacy sanitization.
-
Memory-Efficient Differentially Private Training with Gradient Random Projection
DP-GRAPE reduces memory in differentially private neural network training by using random Gaussian projections on gradients instead of SVD, achieving comparable privacy-utility tradeoffs to DP-SGD and scaling to 6.7B parameter models.
-
Agents That Know Too Much: A Data-Centric Survey of Privacy in LLM Agents
A data-centric survey finds that only information-flow control covers compositional and cross-session leakage in LLM agents and that no single benchmark tests an agent across all its data surfaces under one policy.
-
Selective Token-Level Cryptographic Redaction for Privacy-Preserving Clinical Deployment of Large Language Models
HERALD selectively encrypts sensitive tokens via medical NER, POS policies, and deterministic ciphertext substitution to enable privacy-preserving clinical LLM use while recovering near-plaintext task performance.