aiqbench analysis
This research addresses the critical tension between fine-tuning LLMs for domain adaptation and preventing private data leakage, proposing a novel post-hoc pruning method that minimizes utility loss while significantly reducing PII inference accuracy. By leveraging a joint-embedding predictive architecture to distinguish identity-specific neurons from useful knowledge, JEPA-Prune achieves superior privacy protection with negligible impact on model perplexity compared to existing editing techniques. The findings offer a practical pathway for deploying specialized models in sensitive environments without compromising performance or violating data privacy norms.
Prune the Leaky Neurons: Mitigating Private Data Leakage in Fine-Tuned LLMs via Joint-Embedding-Guided Pruning
arXiv:2610.22792v1 Announce Type: new Abstract: Fine-tuning large language models (LLMs) on proprietary records enables domain adaptation but introduces two challenges: (i) privacy leakage of memorized private data, such as personally identifiable information (PII), and (ii) balancing private data removal with utility preservation. Existing defenses, including data sanitization and differential privacy, often degrade performance, while privacy-neuron editing lacks a PII-free reference to distinguish identity-specific neurons from those supporting useful text. We propose JEPA-Prune, a post-hoc neuron pruning framework guided by a model trained exclusively on PII-free views. Alongside the fine-tuned private model, JEPA-Prune, trains a public model on surrogate and masked views of the same records using a joint-embedding predictive architecture (JEPA), capturing domain knowledge without identifiers. A cross-model representation probe quantifies deviations between private representations of leaked records and this PII-free reference. Integrated gradients and a scale-invariant specificity score then identify PII-responsive neurons, which are pruned without retraining. Experiments show that pruning just 1.6% of neurons reduces PII inference accuracy by up to 7.4x and PII extraction recall by up to 13.9x, while keeping unseen-text perplexity within 16% of that of the unpruned model for most LLMs. In contrast, privacy-neuron editing increases perplexity by at least 26x.
Abstract:Fine-tuning large language models (LLMs) on proprietary records enables domain adaptation but introduces two challenges: (i) privacy leakage of memorized private data, such as personally identifiable information (PII), and (ii) balancing private data removal with utility preservation. Existing defenses, including data sanitization and differential privacy, often degrade performance, while privacy-neuron editing lacks a PII-free reference to distinguish identity-specific neurons from those supporting useful text. We propose JEPA-Prune, a post-hoc neuron pruning framework guided by a model trained exclusively on PII-free views. Alongside the fine-tuned private model, JEPA-Prune, trains a public model on surrogate and masked views of the same records using a joint-embedding predictive architecture (JEPA), capturing domain knowledge without identifiers. A cross-model representation probe quantifies deviations between private representations of leaked records and this PII-free reference. Integrated gradients and a scale-invariant specificity score then identify PII-responsive neurons, which are pruned without retraining. Experiments show that pruning just 1.6% of neurons reduces PII inference accuracy by up to 7.4x and PII extraction recall by up to 13.9x, while keeping unseen-text perplexity within 16% of that of the unpruned model for most LLMs. In contrast, privacy-neuron editing increases perplexity by at least 26x.
Computation and Language (cs.CL); Artificial Intelligence (cs.AI)
arXiv:2610.22792 [cs.CL]
(or arXiv:2610.22792v1 [cs.CL] for this version)
https://doi.org/10.48550/arXiv.2610.22792 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Full text under the publisher's open licence · 290 words
Related coverage
Apple Workshop on Privacy-Preserving Machine Learning & AI 2026
Apple · May 8, 2026
[AINews] Meta Muse Wonder & Wonder Full: MSL’s Claude 5.5-tier Frontier Model
Latent Space · Oct 13, 2026
[AINews] Agents disprove Polysemanticity; greatly improve Interpretability & Alignment
Latent Space · Oct 16, 2026
[AINews] not much happened today
Latent Space · Sep 10, 2026