1. Abstract

Communication-Efficient Federated Learning for Edge Devices is a machine learning project designed to reduce communication overhead during collaborative model training across distributed edge devices. Traditional machine learning systems often require centralized data collection, while standard federated learning repeatedly exchanges model updates between clients and a central server, consuming network bandwidth and energy. The proposed system uses communication optimization techniques such as model update compression, quantization, sparsification, and reduced communication rounds to improve training efficiency. Each edge device trains a local model using its own data and transmits optimized model updates to a central server. The server aggregates these updates to generate a global model and distributes it back to participating devices. This approach aims to reduce communication costs, improve training efficiency, and support scalable federated learning in resource-constrained environments such as IoT networks, mobile devices, and edge computing systems.

2. Objectives

1. To develop a communication-efficient federated learning framework for edge devices.
2. To reduce network bandwidth consumption during model training.
3. To minimize communication rounds between clients and the central server.
4. To implement model update compression and quantization techniques.
5. To enable collaborative learning without directly sharing raw local data.
6. To reduce training latency and communication overhead.
7. To evaluate model accuracy, training time, bandwidth usage, and communication costs.
8. To develop a dashboard for monitoring federated training performance.

3. Existing System

Traditional machine learning systems generally transfer data to a central server for model training. Standard Federated Learning addresses data-sharing concerns by training models locally, but it can require frequent communication between clients and the server. These repeated exchanges can consume significant bandwidth and energy, particularly when edge devices have limited processing power, network connectivity, and battery capacity.

Disadvantages

1. High communication overhead during repeated training rounds.
2. Increased bandwidth consumption.
3. Longer training time in low-bandwidth environments.
4. Higher energy consumption for communication.
5. Limited suitability for resource-constrained edge devices.
6. Performance degradation when network connectivity is unstable.

4. Proposed System

The proposed system combines Federated Learning with communication optimization techniques to improve distributed model training. Each edge device trains a local machine learning model using its own dataset. Before transmitting updates, the client applies techniques such as quantization, sparsification, or update compression to reduce the size of transmitted information. The central server aggregates the updates using Federated Averaging (FedAvg) and sends the global model back to clients.

The system can also use fewer communication rounds, adaptive client participation, or local training for multiple epochs between server updates. It measures communication cost alongside predictive performance to identify an effective balance between accuracy and efficiency.

Key Features

1. Federated learning across multiple simulated edge clients.
2. Model update compression and quantization.
3. Reduced communication frequency.
4. FedAvg-based global model aggregation.
5. Bandwidth and communication cost monitoring.
6. Training time and accuracy comparison.
7. Web-based performance visualization.

5. Implementation Procedure

1. Dataset Collection: Select a suitable dataset for classification or prediction, such as human activity recognition, IoT sensor analysis, or image classification.

2. Data Preprocessing: Handle missing values, remove duplicates, encode categorical variables, and normalize numerical features.

3. Edge Client Simulation: Divide the dataset into separate client partitions representing different edge devices.

4. Local Model Development: Build a machine learning model using Scikit-learn or a neural network using PyTorch or TensorFlow.

5. Federated Learning Setup: Use Flower to coordinate communication between the edge clients and the central server.

6. Baseline Training: Implement standard Federated Learning using FedAvg and record accuracy, communication rounds, update sizes, and training time.

7. Model Update Compression: Apply quantization, sparsification, or other compression methods to reduce the size of client updates.

8. Communication Optimization: Experiment with local training epochs, fewer communication rounds, and suitable client participation strategies.

9. Model Aggregation: Aggregate the client updates at the central server and distribute the updated global model.

10. Performance Evaluation: Compare the optimized approach with baseline Federated Learning using accuracy, total transmitted data, communication rounds, and training time.

11. Dashboard Development: Create a Streamlit or Flask dashboard to display model accuracy, communication costs, and training progress.

12. Testing: Test different client counts and simulated network conditions to assess scalability and robustness.

6. Software Requirements

7. Hardware Requirements

Recommended: Intel Core i5 or equivalent, 16 GB RAM, and SSD storage for smoother experimentation with multiple clients.

8. Advantages of the Project

1. Reduced Communication Cost: Compresses model updates to reduce transmitted data.
2. Lower Bandwidth Usage: Helps federated learning operate in bandwidth-constrained environments.
3. Faster Training Potential: Fewer communication rounds can reduce overall training time.
4. Energy Efficiency Potential: Reduced communication may lower network-related energy consumption.
5. Data Locality: Raw training data remains on individual clients.
6. Scalability: Supports experiments involving multiple edge devices.
7. Flexible Optimization: Enables comparisons of quantization, sparsification, and local training strategies.
8. Performance Monitoring: Measures the trade-off between communication efficiency and model accuracy.