Vertical Federated Learning (VFL) is a privacy-preserving machine learning paradigm where two or more collaborating parties hold datasets with different feature spaces but a significant overlap in the sample space (i.e., the same entities or users). Unlike horizontal federated learning where data is partitioned by rows, VFL addresses scenarios where features are split vertically across silos—for example, a bank holding financial history and an e-commerce platform holding purchase records for the same customers. The primary technical challenge is private entity alignment, which securely identifies common records without exposing non-overlapping identities.
Glossary
Vertical Federated Learning
A federated learning paradigm where collaborating parties hold data with different features about the same set of overlapping entities, requiring entity alignment and split neural network training.
DEFINITION
What is Vertical Federated Learning?
A federated learning paradigm where collaborating parties hold data with different features about the same set of overlapping entities, requiring entity alignment and split neural network training.
Training in VFL typically employs a split neural network architecture, where each party maintains a local bottom model that processes its proprietary features, and a top model aggregates the intermediate representations on a server. During forward propagation, parties exchange encrypted intermediate activations rather than raw data. Backpropagation requires the server to compute and distribute gradients for the top model while each party independently updates its bottom model. This architecture enables collaborative model building for applications like joint credit scoring or multi-institutional biomarker discovery where no single organization possesses all predictive features for a given patient or customer.
ARCHITECTURAL PARADIGM
Key Characteristics of Vertical Federated Learning
Vertical Federated Learning (VFL) addresses the scenario where collaborating parties hold data with different features about the same set of overlapping entities. Unlike horizontal federated learning, VFL requires entity alignment and split neural network training to collaboratively build a model without exposing raw feature data.
01
Entity Alignment
The foundational prerequisite for VFL where parties identify overlapping samples across their silos using Private Set Intersection (PSI) protocols. This cryptographic step matches records belonging to the same entity without revealing non-overlapping entries.
- Uses encrypted identifiers like email hashes or medical record numbers
- Ensures only common entities participate in training
- Prevents data leakage of unique records held by a single party
02
Split Neural Network Architecture
The model is physically partitioned so each party holds a bottom sub-network that processes its local features into intermediate representations called embeddings. These embeddings are exchanged instead of raw data.
- A top sub-network on a server aggregates embeddings to compute the final prediction
- Gradients flow backward through the split, updating local sub-networks
- Raw features never leave their originating institution
03
Asymmetric Feature Ownership
Only one party typically holds the label column (the prediction target), creating an inherent information asymmetry. This party often acts as the active server or label owner.
- Common in finance: bank has credit scores, retailer has purchase history
- Common in healthcare: hospital has diagnosis labels, lab has genomic features
- Requires careful gradient and loss sharing protocols
04
Privacy-Preserving Computation
VFL employs cryptographic techniques to protect intermediate data exchanged during training. Homomorphic Encryption (HE) allows computation on encrypted embeddings, while Differential Privacy (DP) adds calibrated noise to gradients.
- Prevents inference attacks on shared embeddings
- Defends against honest-but-curious adversaries
- Balances privacy budget with model accuracy
05
Inference in Production
At inference time, all parties must collaborate to generate a prediction for a new sample. This requires a real-time orchestration layer that coordinates embedding computation across distributed sub-networks.
- Introduces latency constraints not present in horizontal FL
- Requires high-availability infrastructure at each participating site
- Can use TEEs to accelerate secure inference
06
Gradient Leakage Defense
Shared embeddings and gradients can leak information about private features. VFL implementations must defend against embedding inversion attacks where an adversary reconstructs raw inputs from intermediate representations.
- Gradient compression reduces information leakage
- Adversarial training adds robustness to inversion attempts
- Secure aggregation masks individual contributions
DATA PARTITIONING PARADIGMS
Vertical vs. Horizontal Federated Learning
Structural comparison of the two primary federated learning topologies based on how data is distributed across collaborating parties.
| Feature | Vertical FL | Horizontal FL |
|---|---|---|
Data Partitioning Axis | By feature space (columns) | By sample space (rows) |
Overlapping Entities | Same entities, different features | Different entities, same features |
Entity Alignment Required | ||
Typical Architecture | Split neural network | Full local model training |
Primary Use Case | Cross-industry collaboration (e.g., bank + retailer) | Multi-site clinical trials (e.g., hospital + hospital) |
Communication Pattern | Intermediate activations and gradients | Model weights or gradients |
Privacy Mechanism | Entity alignment via Private Set Intersection | Secure aggregation of updates |
Computational Overhead | Higher (synchronized forward/backward passes) | Lower (independent local training) |
VERTICAL FEDERATED LEARNING
Frequently Asked Questions
Clear, technical answers to the most common questions about entity-aligned, feature-partitioned collaborative model training.
Vertical Federated Learning (VFL) is a privacy-preserving machine learning paradigm where collaborating parties hold data with different features about the same set of overlapping entities. Unlike horizontal federated learning where datasets share the same feature space, VFL addresses scenarios where, for example, a hospital holds lab results and a pharmacy holds prescription records for the same patients. The process requires a critical private entity alignment step using cryptographic protocols like Private Set Intersection (PSI) to identify common entities without revealing non-overlapping records. Training typically employs a split neural network architecture, where each party maintains a local bottom model that processes its own features, and intermediate activations—not raw data—are exchanged to a top model for joint computation. The gradients are then backpropagated to update local parameters, ensuring that no party ever sees another's raw feature values.
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
Read more
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
Read more
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
Read more
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.
LinkedIn
Limited slotsGet a Free AI Consultation
Partnered with leading AI, data, and software stack.
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.
01
Review the use case
We understand the task, the users, and where AI can actually help.
Read more02
Pick the right approach
We define what needs search, automation, or product integration.
Read more03
Build the first useful version
We implement the part that proves the value first.
Read more04
Improve from there
We add the checks and visibility needed to keep it useful.
Read moreThe first call is a practical review of your use case and the right next step.
Talk to Us