Background

The widespread adoption of autonomous vehicles (AVs) and advanced driver assistance systems has dramatically increased the volume of sensor data generated at the network edge. Leveraging this data for intelligent driving services requires machine learning models that are both accurate and privacy conscious. Federated Learning (FL) has emerged as a promising paradigm in this context, enabling model training across distributed vehicles without transferring raw data to centralized servers.

FL allows multiple clients to collaboratively train a global model while keeping their local datasets private, effectively breaking data silos and protecting sensitive information. In a typical FL setup, a central server distributes a global model to multiple vehicles, which then train the model locally on their own sensor data. The vehicles send their updated model parameters back to the server, which aggregates them to update the global model.

A critical challenge in FL arises from communication constraints, particularly in real-world deployments with bandwidth-limited vehicles. Exchanging model updates frequently can create significant overhead, slowing down convergence and limiting scalability. Various strategies have been proposed to reduce communication, such as selective client participation, parameter compression, quantization, and sparsification. Among these, gradient sparsification—transmitting only a subset of gradient components—has been shown to drastically reduce communication while maintaining model performance. In this thesis, the goal is to investigate a novel approach to reducing communication overhead while preserving model performance. By optimizing communication efficiency and selectively transmitting essential parameters, the proposed approach will provide a practical and scalable solution for bandwidth-constrained FL systems in real-world autonomous vehicle networks.

Description

In this master’s thesis project, you will focus on:

• Investigating the challenges of deploying federated learning (FL) in autonomous vehicle networks, with an emphasis on communication efficiency and bandwidth constraints.
• Exploring advanced techniques for communication optimization, including gradient sparsification, model compression, quantization, and intelligent parameter selection.
• Designing and implementing a communication efficient technique to reduce the size of model updates transmitted between vehicles and the central server.
• Evaluating the impact of implemented technique on model accuracy, convergence speed, and communication overhead in both IID and non-IID data settings.
• Comparing the proposed approach with traditional parameter selection methods such as Top-K and Rand-K, assessing trade-offs between communication efficiency and model performance.
• Conducting experiments on ZoD datasets (https://github.com/zenseact/zod) and realistic autonomous vehicle data to assess scalability, adaptability, and robustness of the proposed approach.
• Investigating the privacy and latency benefits of selective parameter transmission in real-world vehicle fleets.

Key Responsibilities

• Develop and implement efficient communication strategies that streamline internal and external information flow, optimize team collaboration, and enhance organizational alignment with company goals.
• Lead and mentor the communications team to ensure consistent messaging across all channels, maintain brand integrity, and manage press, media relations, and stakeholder communications effectively.
• Monitor, analyze, and improve communication processes and tools to maximize clarity, reduce information bottlenecks, and ensure timely and impactful delivery of messages across the organization.

Qualifications

We are looking for highly motivated students with a good general background in machine learning and computer vision. The following skills would be essential:

• Deep learning
• Federated learning (would be a bonus)
• Python programming
• Reading scientific papers
• Handling complex systems

Terms

• Industry supervisor: Sima Sinaei, PhD (sima.sinaei@ri.se)
• Division, department: Digital Systems division, Industrial Systems department
• Location: RISE, Kista, Stockholm
• Application deadline: November 15^th, 2025
• Starting date: As soon as possible, not later than December 1^st, 2025.
• Credits: 30 points
• Compensation: 30 000 SEK upon a successful completion of a high-quality thesis

Welcome with your application!