Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

The Sherpa.ai Blind Vertical Federated Learning Paradigm to Minimize the Number of Communications

Alex Acero, Daniel M. Jimenez-Gutierrez, Dario Pighin, Enrique Zuazua, Joaquin Del Rio, Xabi Uribe-Etxebarria

TL;DR

This paper tackles the heavy communication burden and privacy risks of vertical federated learning (VFL) by introducing Sherpa.ai Blind Vertical Federated Learning (SBVFL). SBVFL replaces direct gradient-based coordination with server-generated synthetic labels, enabling nodes to train independently and drastically reducing interparty communications by about $\sim 99\%$ while maintaining competitive accuracy. The approach is supported by a neural-ODE-based theoretical analysis showing that the federated model can match centralized performance under realistic assumptions, and is reinforced by privacy analyses that protect against node-to-server, server-to-node, and node-to-node attacks. Empirical validation includes EMNIST image classification across seven vertical partitions and a real-world financial default prediction case (bank–insurance), illustrating SBVFL’s practical viability for privacy-critical, bandwidth-constrained domains such as healthcare, finance, and defense. Overall, SBVFL provides a scalable, privacy-preserving alternative to standard VFL with significant reductions in communication overhead and robust performance guarantees.

Abstract

Federated Learning (FL) enables collaborative decentralized training across multiple parties (nodes) while keeping raw data private. There are two main paradigms in FL: Horizontal FL (HFL), where all participant nodes share the same feature space but hold different samples, and Vertical FL (VFL), where participants hold complementary features for the same samples. While HFL is widely adopted, VFL is employed in domains where nodes hold complementary features about the same samples. Still, VFL presents a significant limitation: the vast number of communications required during training. This compromises privacy and security, and can lead to high energy consumption, and in some cases, make model training unfeasible due to the high number of communications. In this paper, we introduce Sherpa.ai Blind Vertical Federated Learning (SBVFL), a novel paradigm that leverages a distributed training mechanism enhanced for privacy and security. Decoupling the vast majority of node updates from the server dramatically reduces node-server communication. Experiments show that SBVFL reduces communication by ~99% compared to standard VFL while maintaining accuracy and robustness. Therefore, SBVFL enables practical, privacy-preserving VFL across sensitive domains, including healthcare, finance, manufacturing, aerospace, cybersecurity, and the defense industry.

The Sherpa.ai Blind Vertical Federated Learning Paradigm to Minimize the Number of Communications

TL;DR

This paper tackles the heavy communication burden and privacy risks of vertical federated learning (VFL) by introducing Sherpa.ai Blind Vertical Federated Learning (SBVFL). SBVFL replaces direct gradient-based coordination with server-generated synthetic labels, enabling nodes to train independently and drastically reducing interparty communications by about while maintaining competitive accuracy. The approach is supported by a neural-ODE-based theoretical analysis showing that the federated model can match centralized performance under realistic assumptions, and is reinforced by privacy analyses that protect against node-to-server, server-to-node, and node-to-node attacks. Empirical validation includes EMNIST image classification across seven vertical partitions and a real-world financial default prediction case (bank–insurance), illustrating SBVFL’s practical viability for privacy-critical, bandwidth-constrained domains such as healthcare, finance, and defense. Overall, SBVFL provides a scalable, privacy-preserving alternative to standard VFL with significant reductions in communication overhead and robust performance guarantees.

Abstract

Federated Learning (FL) enables collaborative decentralized training across multiple parties (nodes) while keeping raw data private. There are two main paradigms in FL: Horizontal FL (HFL), where all participant nodes share the same feature space but hold different samples, and Vertical FL (VFL), where participants hold complementary features for the same samples. While HFL is widely adopted, VFL is employed in domains where nodes hold complementary features about the same samples. Still, VFL presents a significant limitation: the vast number of communications required during training. This compromises privacy and security, and can lead to high energy consumption, and in some cases, make model training unfeasible due to the high number of communications. In this paper, we introduce Sherpa.ai Blind Vertical Federated Learning (SBVFL), a novel paradigm that leverages a distributed training mechanism enhanced for privacy and security. Decoupling the vast majority of node updates from the server dramatically reduces node-server communication. Experiments show that SBVFL reduces communication by ~99% compared to standard VFL while maintaining accuracy and robustness. Therefore, SBVFL enables practical, privacy-preserving VFL across sensitive domains, including healthcare, finance, manufacturing, aerospace, cybersecurity, and the defense industry.
Paper Structure (30 sections, 2 theorems, 40 equations, 10 figures, 3 tables, 1 algorithm)

This paper contains 30 sections, 2 theorems, 40 equations, 10 figures, 3 tables, 1 algorithm.

Table of Contents

  1. Introduction
  2. Motivation and Challenges
  3. Contributions
  4. Related work
  5. Communication Efficiency in VFL
  6. Privacy and Security in VFL
  7. Positioning and Applications
  8. Background and Problem Formulation
  9. Mathematical Description of VFL
  10. Synthetic Label Generation Mechanism
  11. The SBVFL Paradigm
  12. Theoretical Analysis of SBVFL
  13. Privacy and Security Analysis of SBVFL
  14. Node-to-server Attacks
  15. Server-to-node Attacks
  16. ...and 15 more sections

Key Result

Theorem 6.1

Suppose that $d_k\geq 2$ for all $k\in\llbracket P \rrbracket$. Assume that the centralized dataset $\left\{\left(\vec{x}^{\,i}, \vec{y}^{\,i}\right)\right\}_{i=1}^N$ has distinct input data $\vec{x}^{\,i}\neq \vec{x}^{\,j}$ for $i\neq j\in\llbracket N \rrbracket$. Then, the federated model $\emph{F

Figures (10)

  • Figure 1: Graphical description of the HFL and VFL scenarios.
  • Figure 2: Synthetic labels generation through the flow of a hyperbolic PDE.
  • Figure 3: Scheme of the training rounds between nodes and server during VFL.
  • Figure 4: Scheme of the training process for SBVFL.
  • Figure 5: Example of EMNIST Digits image split in vertical slices, to be distributed among the nodes $\{\mathcal{F}_k\}_{k=1}^7$.
  • ...and 5 more figures

Theorems & Definitions (9)

  • Remark 3.1
  • Remark 5.1
  • Theorem 6.1
  • proof
  • Remark 6.1
  • Definition 7.1: $\mathcal{F}_k\leftarrow \mathscr S$ attack
  • Definition 7.2: $\mathcal{F}_k$ security
  • Theorem 7.3
  • proof