---
title: "DP-DiPP: Scalable Differentially Private Data Compression via Diffusion and Stochastic Codes"
source: "arXiv AI/CL/LG（论文）"
source_tier: "t2"
url: "https://arxiv.org/abs/2607.03392v1"
permalink: "https://feed.astralor.com/items/itm_f9db5669d57bc37d413cb6c5"
published: "2026-07-03T14:49:31.000Z"
score: 79
channel: "ai"
tags: ["DP-DiPP","Differential Privacy","Diffusion Models","Data Compression","Poisson Private Representation","Stochastic Codes"]
---

# DP-DiPP: Scalable Differentially Private Data Compression via Diffusion and Stochastic Codes

## AI commentary

> What makes this paper stand out is its clever fusion of diffusion models' generative power with stochastic codes' privacy properties, achieving over 10x compression improvement. For AI researchers and engineers working on privacy-preserving computing, data compression, and image processing, DP-DiPP offers a directly applicable framework that could address real-world bottlenecks in storing and transmitting high-dimensional private data.

## At a glance

This paper proposes DP-DiPP, a compression pipeline combining diffusion models with stochastic codes to address the challenge of compressing high-dimensional image data under differential privacy. It provides direct control over the compression rate-privacy-utility tradeoff, extends the Poisson Private Representation (PPR) to encode privacy mechanism outputs, and integrates with DiffC diffusion-based lossy compression. Experiments on CIFAR-10 demonstrate a 10-30x improvement in compression ratio over the baseline while maintaining comparable privacy guarantees and utility.

## Key points

- Proposes DP-DiPP, the first compression pipeline combining diffusion models with stochastic codes for differentially private image compression.
- Extends Poisson Private Representation (PPR) to encode privacy mechanism outputs with theoretical guarantees.
- Achieves 10-30x better compression ratio on CIFAR-10 image classification compared to baseline.
- Allows flexible trade-off control among compression rate, privacy, and utility.

## Full text

The ever-increasing collection of personal data has created mounting pressure to develop technologies that protect sensitive aspects of individual identity. Differential privacy (DP) provides a principled framework with strong formal guarantees and has already achieved practical success. However, releasing high-dimensional data, such as images, has remained elusive: releasing uncompressed privatized data requires significant storage. At the same time, no effective data compression scheme exists that can compress high-resolution data with privacy guarantees. We address this challenge with DP-DiPP, a compression pipeline that combines stochastic codes with diffusion models. DP-DiPP is highly flexible: the practitioner has direct control over the compression rate-privacy-utility tradeoff. As the theoretical backbone, we extend the Poisson private representation (PPR) to encode the outputs of privacy mechanisms. We then combine it with DiffC, a diffusion-based lossy data compression method, to obtain a differentially private image compressor. Our experiments on privatized image classification on CIFAR-10 demonstrate that DP-DiPP significantly outperforms the baseline, achieving a 10-30 times better compression while retaining comparable privacy guarantees and utility.

---

Verify against the original before citing; curated by Astral Feed: https://feed.astralor.com/items/itm_f9db5669d57bc37d413cb6c5