Differentially Private Motif-Preserving Multi-modal Hashing

2026-05-14

AI Summary

This work addresses the vulnerability of user interaction–derived semantic graphs in cross-modal hashing to link reconstruction attacks and the challenge of preserving essential relational structures under edge differential privacy. The authors propose DMP-MH, a framework employing a Sanitize-then-Distill strategy: it first controls triangle motif sensitivity via deterministic node-degree clipping, identifies and formally names the “Hubness Explosion” problem, and establishes a data-size–independent upper bound on sensitivity; then generates an edge-differentially private synthetic graph using Noisy Mirror Descent; and finally aligns modalities through a dual-stream hashing network with a structured distillation loss. Experiments on MIRFlickr-25K and NUS-WIDE show that under strict privacy guarantees, the method improves mAP by up to 11.4 points over existing baselines while retaining 92.5% of the non-private model’s performance.

Abstract

Cross-modal hashing enables efficient retrieval by encoding images and text into compact binary codes. State-of-the-art methods rely on semantic similarity graphs derived from user interactions for supervision, yet these graphs encode sensitive behavioral patterns vulnerable to link reconstruction attacks. Existing privacy-preserving approaches fail on graph-structured data: Differentially Private SGD destroys relational motifs by treating samples independently, while graph synthesis methods suffer from unbounded local sensitivity in scale-free networks, hub nodes cause single-edge modifications to alter triangle counts by $\mathcal{O}(N)$, necessitating prohibitive noise injection. We term this phenomenon Hubness Explosion. We propose DMP-MH, a Sanitize-then-Distill framework that decouples privacy from representation learning. Our approach first bounds sensitivity by deterministically clipping node degrees, capping the $L_2$-sensitivity of triangle motifs independently of dataset size. A sanitized synthetic graph is then generated via Noisy Mirror Descent under $(\epsilon,\delta)$-Edge Differential Privacy. Finally, dual-stream hashing networks distill this topology using a holistic structural loss that enforces cross-modal alignment. Evaluated on MIRFlickr-25K and NUS-WIDE under a strict inductive protocol, DMP-MH outperforms private baselines by up to 11.4 mAP points while retaining up to 92.5% of non-private performance.

Problem

• Differential Privacy
• Cross-modal Hashing
• Graph Sensitivity
• Motif Preservation
• Link Reconstruction Attack

Innovation

• Differentially Private Hashing
• Motif Preservation
• Hubness Explosion
• Cross-modal Retrieval
• Graph Sanitization