Glossary

Vertical Federated Learning

A federated learning paradigm where participants hold different feature spaces for the same sample population, requiring entity alignment and privacy-preserving feature intersection to train a joint model.

Get in touch Learn more

MLOps engineer reviewing model serving infrastructure on laptop, container orchestration visible, technical workspace.

FEATURE-PARTITIONED COLLABORATION

What is Vertical Federated Learning?

Vertical Federated Learning (VFL) is a privacy-preserving machine learning paradigm where two or more parties hold different feature spaces for an overlapping set of samples, enabling collaborative training of a joint model without exposing raw data. Unlike horizontal federated learning, which partitions data by samples, VFL partitions by attributes, requiring secure entity alignment to identify common records across silos before training begins.

During VFL training, each party maintains its own bottom model for local feature extraction, while a top model aggregates intermediate representations using privacy-preserving techniques such as homomorphic encryption or secure multi-party computation. This architecture allows a bank and an e-commerce platform to jointly train a credit scoring model—each contributing distinct user attributes—without revealing individual customer records to the other party.

ARCHITECTURE PRIMER

Key Characteristics of VFL

Vertical Federated Learning (VFL) is a paradigm shift for organizations holding different features about the same users. Unlike horizontal FL, VFL requires cryptographic entity alignment and privacy-preserving feature intersection before a joint model can be trained.

Entity Alignment

The foundational step in VFL where parties identify overlapping samples without exposing non-matching records. This typically uses Private Set Intersection (PSI) protocols.

Objective: Match users across silos without revealing unique identities.
Mechanism: Cryptographic hashing or homomorphic encryption.
Challenge: Handling fuzzy matching (e.g., typos in names) under encryption is computationally expensive.

PSI

Core Protocol

Split Neural Network Architecture

VFL often employs a split learning topology where the model is physically partitioned across participants.

Bottom Model: Each party trains a local feature extractor on their private features.
Top Model: A server aggregates intermediate representations (embeddings) to compute the final loss.
Cut Layer: The point where local outputs are transmitted; gradients flow back without exposing raw features.

Embeddings

Shared Artifact

Privacy-Preserving Gradient Exchange

VFL must protect against gradient leakage attacks where an honest-but-curious server could reconstruct private features from transmitted embeddings.

Defense: Apply Differential Privacy (DP) noise to intermediate outputs.
Encryption: Use Homomorphic Encryption (HE) to allow the server to compute on encrypted embeddings.
Trade-off: Noise addition reduces model accuracy but provides formal privacy guarantees.

ε < 8

Privacy Budget

Asymmetric Feature Ownership

In a typical VFL scenario, one party often holds the label (e.g., a bank with loan default status) while others hold predictive features (e.g., a retailer with purchase history).

Label Owner: Usually the active party initiating the training.
Feature Owners: Passive parties contributing attributes.
Incentive: Feature owners must be compensated or assured of zero label leakage.

1 Label Owner

Typical Topology

Inference Phase Coordination

Unlike horizontal FL where the final model is distributed, VFL inference requires real-time collaboration between all parties.

Online Serving: All feature owners must be available to run their bottom models.
Latency: Network overhead between parties can introduce significant prediction delays.
Solution: Pre-computed embeddings or secure hardware enclaves to accelerate serving.

< 100ms

SLA Target

Semi-Honest Security Model

Most VFL protocols assume honest-but-curious participants who follow the protocol correctly but may attempt to infer private data from received messages.

Threat: Feature inference attacks via model inversion.
Mitigation: Secure Multi-Party Computation (SMPC) for aggregation.
Limitation: Malicious adversaries sending poisoned embeddings require robust aggregation rules.

SMPC

Defense Standard

VERTICAL FEDERATED LEARNING

Frequently Asked Questions

Clear answers to the most common technical questions about training machine learning models across vertically partitioned data silos.

Vertical Federated Learning (VFL) is a privacy-preserving machine learning paradigm where multiple parties hold different feature spaces for the same set of overlapping entities, and collaboratively train a model without exposing their raw data. Unlike horizontal federated learning where parties share the same features for different samples, VFL addresses scenarios where, for example, a bank holds financial transaction data and an e-commerce platform holds purchase history for the same customers. The process begins with Private Set Intersection (PSI) to identify the common sample population without revealing non-overlapping entities. Once aligned, each party trains a local bottom model on their respective features, and the intermediate representations (embeddings) are exchanged for joint computation of the final prediction. The key technical challenge is that the loss computation and gradient propagation require information from all parties, necessitating secure multi-party computation or homomorphic encryption to protect intermediate values during the collaborative forward and backward passes.

Enabling Efficiency, Speed & Accuracy

Intelligent Analysis, Decision & Execution

We build AI systems for teams that need search across company data, workflow automation across tools, or AI features inside products and internal software.

Talk to Us

Search across company data

Give teams answers from docs, tickets, runbooks, and product data with sources and permissions.

Useful when people spend too long searching or get different answers from different systems.

Enterprise searchRAGPermissions

Automate internal workflows

Use AI to route work, draft outputs, trigger actions, and keep approvals and logs in place.

Useful when repetitive work moves across multiple tools and teams.

AI agentsWorkflow automationGovernance

Add AI to products and internal tools

Build assistants, guided actions, or decision support into the software your team or customers already use.

Useful when AI needs to be part of the product, not a separate tool.

AI integrationDecision supportModel routing

About the author

Prasad Kumkar

CEO & MD, Inference Systems

Prasad Kumkar is the CEO & MD of Inference Systems and writes about AI systems architecture, LLM infrastructure, model serving, evaluation, and production deployment. Over 5+ years, he has worked across computer vision models, L5 autonomous vehicle systems, and LLM research, with a focus on taking complex AI ideas into real-world engineering systems.

His work and writing cover AI systems, large language models, AI agents, multimodal systems, autonomous systems, inference optimization, RAG, evaluation, and production AI engineering.

Limited slotsGet a Free AI Consultation

How We Work

Custom AI workflows for your Business

One-fit-all AI don't work for modern businesses. At Inferensys, we aim to understand your business & custom requirements; which we use to define most efficient agentic workflows, the data, and the tools for your business.

Talk to Us

Vertical Federated Learning

What is Vertical Federated Learning?