TextCrafter: Optimization-Calibrated Noise for Defending Against Text Embedding Inversion
Fuente:
arXiv
Saved in:
| Main Authors: | , , |
|---|---|
| Format: | Preprint |
| Published: |
2025
|
| Subjects: | |
| Online Access: | |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866912839638188032 |
|---|---|
| author | Tang, Duoxun Jiang, Xinhang Niu, Jiajun |
| author_facet | Tang, Duoxun Jiang, Xinhang Niu, Jiajun |
| contents | Text embedding inversion attacks reconstruct original sentences from latent representations, posing severe privacy threats in collaborative inference and edge computing. We propose TextCrafter, an optimization-based adversarial perturbation mechanism that combines RL learned, geometry aware noise injection orthogonal to user embeddings with cluster priors and PII signal guidance to suppress inversion while preserving task utility. Unlike prior defenses either non learnable or agnostic to perturbation direction, TextCrafter provides a directional protective policy that balances privacy and utility. Under strong privacy setting, TextCrafter maintains 70 percentage classification accuracy on four datasets and consistently outperforms Gaussian/LDP baselines across lower privacy budgets, demonstrating a superior privacy utility trade off. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2509_17302 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | TextCrafter: Optimization-Calibrated Noise for Defending Against Text Embedding Inversion Tang, Duoxun Jiang, Xinhang Niu, Jiajun Cryptography and Security Text embedding inversion attacks reconstruct original sentences from latent representations, posing severe privacy threats in collaborative inference and edge computing. We propose TextCrafter, an optimization-based adversarial perturbation mechanism that combines RL learned, geometry aware noise injection orthogonal to user embeddings with cluster priors and PII signal guidance to suppress inversion while preserving task utility. Unlike prior defenses either non learnable or agnostic to perturbation direction, TextCrafter provides a directional protective policy that balances privacy and utility. Under strong privacy setting, TextCrafter maintains 70 percentage classification accuracy on four datasets and consistently outperforms Gaussian/LDP baselines across lower privacy budgets, demonstrating a superior privacy utility trade off. |
| title | TextCrafter: Optimization-Calibrated Noise for Defending Against Text Embedding Inversion |
| topic | Cryptography and Security |
| url | https://arxiv.org/abs/2509.17302 |