Communication- and Computation-Aware Federated Learning for Vehicular Networks
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing machine learning models for vehicular networks face challenges due to high mobility, communication costs, data privacy concerns, and heterogeneity, leading to non-robust predictions and inefficiencies in training and deployment.
Innovation Solution
A communication and computation aware distributed machine learning platform using federated learning, where a centralized learning server coordinates model training across vehicles via roadside units, allowing vehicles to train models locally and share them through a hierarchical communication network, optimizing model distribution and training times based on vehicle-specific computation and data heterogeneity.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of energy
If vehicles train independent machine learning models using local data, then data privacy is protected and communication cost is reduced, but model robustness and prediction accuracy deteriorate due to data imperfection and insufficiency
Solution Approach 1:
The patent merges multiple vehicles' machine learning models through a distributed federated learning framework. Vehicles collaboratively train a global model by sharing model parameters and gradients rather than raw data, combining the computational power and data diversity of multiple vehicles to create a robust model that maintains privacy protection while achieving high prediction accuracy.
2Measurement precision
If vehicles transfer collected data to a central server for centralized training, then model accuracy can be improved through comprehensive data, but communication bandwidth requirement and data privacy threats increase enormously
Solution Approach 1:
The patent introduces a central server as an intermediary that coordinates federated learning without directly storing or processing vehicle data. The server distributes the global model to vehicles, collects trained local models and gradients, and aggregates them to update the global model, thereby enabling accurate model training while maintaining data privacy through encrypted parameter sharing.
3Productivity
If a hierarchical communication network with roadside units is introduced to coordinate distributed training, then model training coordination and aggregation are improved, but device complexity and communication overhead increase
Solution Approach 1:
The patent segments the communication network into hierarchical layers with roadside units (RSUs) handling local coordination and aggregation tasks, while a central server performs global model management. This segmentation distributes computational load and communication overhead across multiple levels, improving training efficiency while managing complexity through modular architecture.
4Reliability
If distributed machine learning is implemented in vehicular networks, then data privacy is protected and communication cost is reduced, but training delay increases due to high mobility and limited connection time
Solution Approach 1:
The patent implements preliminary actions by pre-establishing communication connections with roadside units before vehicles enter coverage areas, and by pre-synchronizing model training schedules. The system predicts vehicle trajectories and connection durations to proactively allocate communication resources and adjust training iterations, reducing actual training delay while maintaining privacy protection.
Data Source
AI summary
A computer-implemented method is provided for training a global machine learning model using a learning server and a set of vehicle agents connected to roadside units (RSUs). It involves selecting agents, associating them with RSUs based on proximity, and transmitting training data and deadlines. Agents train models locally, which are then aggregated through RSUs to refine the global model to desired precision. The method includes steps of selecting vehicle agents from a pool of the vehicle agents connected to the RSUs, associating the selected vehicle agents and the RSUs respectively based on distances from the selected vehicle agents to the RSUs configured to provide measurements of the distances to the learning server, and transmitting a global model, a selected agent set and deadline thresholds in each global training round to the RSUs configured to transmit the global model and training deadlines to the selected vehicle agents.