Federated Learning & Privacy-Preserving AI InfrastructureSobhan Bahrami
Design and implement production Federated Learning (FL) architectures for decentralized edge nodes and strict data privacy compliance (GDPR, HIPAA, EU AI Act).
Specializing in Flower (flwr), NVIDIA FLARE, and PyTorch setups where client data never leaves local devices or regional VPCs.
Core Capabilities:
Decentralized Training & Aggregation: FedAvg, FedProx, and custom adaptive aggregation handling Non-IID client drift.
Privacy Budgeting & Security: Differential Privacy (DP-SGD via Opacus) with explicit (ε, δ) privacy accounting, and Secure Aggregation (SecAgg).
Byzantine Fault Resilience: Fault-tolerant consensus algorithms (Krum, Coordinate-wise Median, Trimmed Mean) defending against poisoned gradient updates.
Edge Bandwidth Optimization: Gradient quantization (4-bit/8-bit) and Top-k sparsification for constrained network environments.
FAQs
Starting at$200 /hr
Duration3 weeks
Tags
Python
PyTorch
AI Engineer
DevOps Engineer
Cybersecurity Engineer
Machine Learning Engineer
Service provided by
Sobhan Bahrami Budapest, Hungary
- 6
- Followers
Federated Learning & Privacy-Preserving AI InfrastructureSobhan Bahrami
Starting at$200 /hr
Duration3 weeks
Tags
Python
PyTorch
AI Engineer
DevOps Engineer
Cybersecurity Engineer
Machine Learning Engineer
Design and implement production Federated Learning (FL) architectures for decentralized edge nodes and strict data privacy compliance (GDPR, HIPAA, EU AI Act).
Specializing in Flower (flwr), NVIDIA FLARE, and PyTorch setups where client data never leaves local devices or regional VPCs.
Core Capabilities:
Decentralized Training & Aggregation: FedAvg, FedProx, and custom adaptive aggregation handling Non-IID client drift.
Privacy Budgeting & Security: Differential Privacy (DP-SGD via Opacus) with explicit (ε, δ) privacy accounting, and Secure Aggregation (SecAgg).
Byzantine Fault Resilience: Fault-tolerant consensus algorithms (Krum, Coordinate-wise Median, Trimmed Mean) defending against poisoned gradient updates.
Edge Bandwidth Optimization: Gradient quantization (4-bit/8-bit) and Top-k sparsification for constrained network environments.
FAQs
$200 /hr