List of Topics:
Location Research Breakthrough Possible @S-Logix pro@slogix.in

Office Address

Social List

Federated Learning on AWS IoT for Edge Devices

Edge Devices

Federated Learning on AWS IoT for Edge Devices

  • Use Case: In many IoT applications (healthcare wearables, smart homes, autonomous vehicles), data privacy and bandwidth efficiency are critical. Instead of sending raw data to the cloud, federated learning enables local training of models on IoT devices and only shares model updates with a central aggregator.This enhances data privacy, reduces network usage, and enables near real-time AI inference at the edge.

Objective

  • Enable collaborative model training across distributed IoT devices without raw data sharing.

    Optimize latency, bandwidth consumption, and model accuracy.

    Propose QoE-aware aggregation algorithms that balance accuracy, real-time inference needs, and user/device constraints (battery, network).

    Demonstrate scalability in heterogeneous IoT environments.

Project Description

  • IoT Device Setup:

    Multiple IoT devices (e.g., Raspberry Pi, sensors) run AWS IoT Greengrass to support ML inference and partial training.

    Each device collects data (temperature, health, motion, etc.) and trains a local ML model.

Federated Learning Workflow

  • Local training occurs on devices.

    Model updates (gradients/weights) are sent securely to AWS S3.

    AWS SageMaker acts as the central server for aggregation of model updates.

    The aggregated global model is redistributed back to devices.

QoE-aware Aggregation Algorithm

  • Devices with poor bandwidth/low battery contribute less frequently.

    Prioritize updates from high-quality, reliable nodes.

    Adaptive aggregation balances accuracy vs. efficiency.

Evaluation Metrics

  • Training latency vs. model accuracy.

    Bandwidth savings compared to centralized training.

    Privacy level (sensitive raw data never leaves device).

    QoE metrics (device resource usage, user experience).
  • AWS Services & Purpose :
    AWS Service Purpose
    AWS IoT Greengrass Deploy ML inference models on IoT devices.
    Allow partial local training.
    Manage edge device connectivity with the cloud.
    Amazon SageMaker Act as the central federated learning server.
    Aggregate model updates from multiple devices.
    Train/update global ML models and redeploy them to devices.
    Amazon S3 Store local model updates securely.
    Maintain version control of models.
    Enable intermediate data exchange between devices and the cloud aggregator.
    AWS CloudWatch Monitor training latency.
    Track bandwidth usage.
    Collect device resource utilization metrics for optimization.