SMAP: A Joint Dimensionality Reduction Scheme for Secure Multi-Party Visualization

Jiazhi Xia, Tianxiang Chen, Lei Zhang, Wei Chen, Yang Chen, Xiaolong Zhang, Cong Xie, Tobias Schreck

Abstract

Nowadays, as data becomes increasingly complex and distributed, data analyses often involve several related datasets that are stored on different servers and probably owned by different stakeholders. While there is an emerging need to provide these stakeholders with a full picture of their data under a global context, conventional visual analytical methods, such as dimensionality reduction, could expose data privacy when multi-party datasets are fused into a single site to build point-level relationships. In this paper, we reformulate the conventional t-SNE method from the single-site mode into a secure distributed infrastructure. We present a secure multi-party scheme for joint t-SNE computation, which can minimize the risk of data leakage. Aggregated visualization can be optionally employed to hide disclosure of point-level relationships. We build a prototype system based on our method, SMAP, to support the organization, computation, and exploration of secure joint embedding. We demonstrate the effectiveness of our approach with three case studies, one of which is based on the deployment of our system in real-world applications.

Introduction

In fields like healthcare or finance, data often exists in "isolated islands." Multiple organizations might want to see how their datasets compare to a global context, but privacy regulations like GDPR and the sensitivity of raw data prevent centralized aggregation. Standard dimensionality reduction tools like t-SNE typically require gathering all raw data on a single server. Even with anonymization, the resulting distance matrix can be used to reverse-engineer sensitive attributes.

SMAP introduces a secure, multi-party framework for t-SNE, allowing different organizations to visualize their combined datasets in a shared space without ever revealing their raw data to one another. By leveraging additive homomorphic encryption and a dual-collaborator architecture, the system produces a joint embedding that is mathematically identical to standard t-SNE while maintaining strict privacy guarantees.

Methods: Secure Multi-Party Computation for t-SNE

The authors reformulate the standard t-SNE algorithm into a Secure Multi-Party Computation (SMPC) protocol. The architecture relies on two "semi-honest," non-colluding collaborators (S and T):

  1. Additive Homomorphic Encryption (AHE): Participants encrypt their data before transmission. AHE allows collaborators to perform mathematical operations, such as addition, on encrypted values without decryption.
  2. The Dual-Collaborator Protocol: Collaborator T receives encrypted data and adds random noise. Collaborator S, holding the decryption key, decrypts this noised data to compute distances. Due to a mathematical "fraction reduction" in the t-SNE probability formula, the noise added by T cancels out during the final computation. This ensures S only sees noised distances, and T only sees encrypted data.
  3. Privacy-Aware Visualization: SMAP offers two levels of visualization. For lower privacy needs, it provides standard scatterplots. For stricter requirements, it uses density maps (heatmaps) to show global distributions without revealing individual data points.

Evaluation and Results

The researchers evaluated SMAP using three distinct scenarios: the Clothing 1M fashion image dataset, the MNIST handwritten digit dataset, and real-world sensitive health data from three community hospitals.

Comparison of joint embeddings for the Clothing 1M dataset: (a) standard t-SNE, (b) and (c) SMAP results for different data parties.

A key result is that SMAP achieves exact accuracy. Unlike many privacy-preserving methods that add "differential privacy" noise which degrades results, SMAP’s protocol ensures the noise cancels out. The resulting visualization is mathematically identical to a standard, non-secure t-SNE.

In the hospital case study, users identified that while their datasets covered similar clinical areas, the density distributions differed significantly (e.g., one hospital had a much younger patient demographic), leading to better collaborative insights. While secure computation is computationally intensive, the authors demonstrated that using a high-performance cluster (300 CPUs) can reduce distance calculations from 32 hours to roughly 29 seconds.

Discussion and Future Work

A major insight of this work is that the distance matrix itself is a privacy risk. By ensuring no single party ever sees the full, accurate distance matrix, SMAP closes a common leak in distributed machine learning. Furthermore, the paper argues that visualization design—such as switching to density maps—is a critical part of the security stack to prevent "point-level" relationship leaks.

The primary limitation remains computational overhead; encryption and decryption are time-consuming for large datasets on standard hardware. Additionally, the current implementation is tailored for t-SNE. Future work aims to optimize encryption speed and explore "secure collaborative visual analysis," allowing users to interactively explore data in a crowdsourced, privacy-preserving manner.

Summary updated

Abstract

Nowadays, as data becomes increasingly complex and distributed, data analyses often involve several related datasets that are stored on different servers and probably owned by different stakeholders. While there is an emerging need to provide these stakeholders with a full picture of their data under a global context, conventional visual analytical methods, such as dimensionality reduction, could expose data privacy when multi-party datasets are fused into a single site to build point-level relationships. In this paper, we reformulate the conventional t-SNE method from the single-site mode into a secure distributed infrastructure. We present a secure multi-party scheme for joint t-SNE computation, which can minimize the risk of data leakage. Aggregated visualization can be optionally employed to hide disclosure of point-level relationships. We build a prototype system based on our method, SMAP, to support the organization, computation, and exploration of secure joint embedding. We demonstrate the effectiveness of our approach with three case studies, one of which is based on the deployment of our system in real-world applications.
BibTeX
@inproceedings{xia2020smap,
  author = {Jiazhi Xia and Tianxiang Chen and Lei Zhang and Wei Chen and Yang Chen and Xiaolong Zhang and Cong Xie and Tobias Schreck},
  title = {{SMAP: A Joint Dimensionality Reduction Scheme for Secure Multi-Party Visualization}},
  booktitle = {IEEE VAST@IEEE VIS},
  year = {2020},
  url = {https://openreview.net/forum?id=0fVzQjKmPLz},
}
Figure