Our Services

Federated learning applied in the real world

Two end-to-end deployments — one driven by regulation, one by physical scale. Both demonstrating that collaborative AI is possible without moving a single byte of raw data.

01
Banking · Combined FL Cross-institutional fraud detection across four institutes
Neural Net Combined FedAvg DP ε=0.8 Flower
02
Astronomy · Horizontal FL Distributed galaxy classification across three survey institutes
CNN FedProx Horizontal Flower

Case Study 01 · Banking · Combined FL

Cross-institutional fraud detection
across four institutes

Fraud rings rarely confine themselves to a single institution. This case study explores how four major financial organisations
built a shared fraud detection model using FL — without any institution exposing its customer data to the others.


01 · Background

The data landscape

Four financial institutions jointly aim to detect fraud, but their data is both horizontally and vertically partitioned. Some customers are shared across institutions, while each organisation observes a different subset of features.

Customer distribution (overlap)

  • Retail Bank — customers C1–C600
  • Corporate Bank — customers C400–C1200
  • Credit Bureau — customers C200–C800
  • Digital Bank — customers C700–C1200

→ Overlapping IDs (e.g., C400–C600) appear in multiple banks, but no institution sees the full population (C1–C1200).

Feature distribution (by institution)

  • Retail Bank — transaction amounts, timestamps, merchant categories
  • Corporate Bank — corporate transfers, account balances, cash flow patterns
  • Credit Bureau — credit scores, loan history, default indicators
  • Digital Bank — device IDs, IP addresses, login behaviour
  • Fraud labels are available only at Retail and Corporate Banks.

→ Features are disjoint across institutions; no single bank has a complete feature vector.


02 · The Problem

Why data sharing is not an option

Each institution holds sensitive financial data governed by strict privacy and banking regulations. Centralising this data — even in encrypted form — would require complex approvals and is often not feasible. Meanwhile, fraud losses are significant, with a large share driven by cross-institution activity. The need for collaboration is clear, but raw data sharing is not viable.


04 · FL Assessment and Design

Technical configuration

ML Model Neural Net 4-layer MLP with dropout — handles sparse labels and mixed feature types.
Partitioning Combined Overlapping customers, different feature sets across all four institutions.
Privacy DP (ε=0.8) Gaussian noise added to gradients before upload.
Aggregation FedAvg Weighted by dataset size per round.
Framework Flower Flower was selected for its framework-agnostic client API and proven cross-silo support.

05 · Implementation

How it was built

01
Entity alignment — private set intersection

Before training, a private set intersection protocol identified shared customer IDs across institutions without revealing the IDs themselves. This produced a shared index used to align vertical features without centralising data.

02
Training — 40 federated rounds

Each round: local training on institution data → DP noise injection → gradient or intermediate information upload to server → FedAvg aggregation → global model broadcast. No raw data left any institution at any point.

03
Validation — held-out test sets per institution

Each institution evaluated the global model on its own labelled holdout set. Results were aggregated by the server into a single performance report — again without sharing institution-level predictions.


06 · Outcomes

Results across the three service pillars

The federated model outperformed each institution's siloed baseline on every metric — demonstrating that collaboration without data sharing is not only legally viable but technically superior.

+34% improvement in cross-institution fraud detection rate vs individual institution baselines
−41% reduction in false positive rate — fewer legitimate transactions incorrectly flagged
100% compliance and privacy maintained — zero raw data shared between institutions

Case Study 02 · Astronomy · Horizontal FL

Distributed galaxy classification
across three astronomical survey institutes

Modern sky surveys generate petabytes of imaging data that no single institution can centralise or process alone. This case study explores how three major astronomical research institutes trained a shared galaxy classification model using horizontal FL — keeping data local where the compute lives.


01 · Background

The data landscape

Three institutes independently operate large-scale sky survey telescopes, each accumulating imaging data at a rate that outpaces network transfer capacity. Each institute captures the same kinds of objects — galaxies, nebulae, stellar clusters — using the same feature schema and labelling conventions. The data is horizontally partitioned: same features, entirely different sky regions and samples.

Data holders

  • Node 1: Imaging data of size 1.8 PB
  • Node 2: Imaging data of size 2.3 PB
  • Node 3: Imaging data of size 1.4 PB

Data types

  • Image — multi-band telescope imaging (RGB + infrared)
  • Labels — galaxy morphology classes (spiral, elliptical, irregular)

02 · The Problem

Why centralisation is physically impossible

Unlike most FL scenarios, the barrier here is not legal or regulatory — it is purely physical. The combined dataset across the three institutes exceeds 5.5 petabytes of raw imaging data. Transferring this volume over even a dedicated 100 Gbps research network would take over 120 days of continuous transfer, with storage and preprocessing costs in the tens of millions of dollars. Furthermore, no single compute facility has the GPU memory capacity to train a CNN on the full combined dataset in memory. The data must stay where the storage and local compute infrastructure already exists.


03 · FL Assessment and Design

Technical configuration

ML Model CNN ResNet-50 backbone fine-tuned for galaxy morphology — pretrained on ImageNet.
Partitioning Horizontal Same feature schema and label set — different sky regions and samples at each institute
Privacy None required Astronomical imaging contains no personal data — standard model updates with no noise addition.
Aggregation FedProx Proximal term (μ=0.01) handles unequal dataset sizes and partial round participation.
Framework Flower Flower was selected for its framework-agnostic client API.

04 · Implementation

How it was built

01
Data preprocessing — local standardisation pipelines

Each institute applied an identical preprocessing pipeline locally: image normalisation to a shared photometric scale, augmentation (random crop, flip, rotation) and resizing to 224×224 pixels. Alignment was verified by comparing summary statistics — not by sharing images — before the first training round.

02
Training — 60 federated rounds with async participation

Each round: local CNN training for 3 epochs on institute GPU cluster → weight upload to CERN server (~280 MB per client) → FedProx aggregation → global model broadcast. Node 1, with fewer GPUs, was permitted to skip up to 20% of rounds without destabilising convergence — a key advantage of FedProx over FedAvg.

03
Validation — cross-institute held-out sky patches

Each institute held back 10% of labelled images for local validation. The global model was also evaluated on a small jointly-curated benchmark set of 12,000 images — the only data physically transferred during the entire project, at under 20 GB.


05 · Outcomes

Results across the three service pillars

The federated CNN matched the accuracy of a hypothetical centralised model trained on all data — without a single raw image leaving its institute of origin. Training that would have required 120+ days of data transfer was completed in under two weeks of federated rounds.

94.2% Galaxy morphology classification accuracy — within 0.8% of a simulated centralised baseline
11 days total training time across 60 rounds — vs 120+ days data transfer for any centralised approach
≈ 280 MB data transferred per round per node — vs petabytes for centralised training

Start your journey toward privacy-preserving AI

Whether you're exploring FL or ready to deploy, we can help you assess, design and implement the right solution for your organisation.

Book a consultation