Master’s thesis “Decentralised federated learning for semantic communications in IIoT Edge-Cloud environments”

Semantic communications shift the goal of data transmission from bit-accurate delivery to task-relevant meaning  reducing bandwidth and latency for industrial applications by transmitting only what matters for a given inference or control task. Training the deep learning models that underpin this paradigm in distributed IIoT environments raises open challenges: data is heterogeneous across devices, raw data cannot be centralised for privacy and bandwidth reasons, and semantic models must adapt to the heterogeneous Edge-Cloud and to channel conditions that vary dynamically across IoT devices, edge nodes, and cloud infrastructure.

Existing semantic communication models are fundamentally local optimisations. In multi-hop edge-cloud deployments, semantic distortion accumulates non-linearly across paths and depends on dynamic interactions between flows and routing decisions that are themselves unaware of semantic content. Federated learning introduces a further complication: aggregation strategies designed for homogeneous settings degrade under the device and data heterogeneity typical of IIoT deployments, and no systematic evaluation of aggregation strategy impact on semantic task performance currently exists.

The thesis investigates decentralised federated learning approaches to train and adapt semantic encoder-decoder models across heterogeneous IIoT edge-cloud environments.

Your tasks:

• Survey and evaluate simulation environments ( ns-3, ns3-ai, ns3-gym, PyTorch) for modelling semantic communication channels and federated learning across IoT-edge-cloud topologies, building on the ns-3 semantic channel module developed as shared infrastructure within SemComIIoT.
• Design and implement a decentralised FL architecture for training semantic encoder-decoder models across a representative IIoT edge-cloud scenario, explicitly addressing data heterogeneity, communication overhead, and model convergence under non-IID conditions.
• Integrate context-awareness mechanisms supporting adaptive semantic model selection and federated aggregation strategy, e.g.,  conditioned on channel SNR, device capability, and task requirements, drawing on the benchmarking results from the parallel FL aggregation study (FedAvg vs FedProx vs personalised FL variants).
• Evaluate the achieved framework via ns-3 simulations. Document the design and contribute results to open tools and publications within SemComIIoT.

__________

Your profile:

• Master student in Computer Science, Electrical Engineering, Communications Engineering, or a related field.
• Background in machine learning, federated learning, and/or distributed systems (knowledge of semantic communications, edge AI, or IIoT architectures is a plus).
• Practical programming experience in Python, Java, C/C++, or similar.
• Familiarity with ML/FL frameworks (e.g., PyTorch, Flower, FedML) and/or network simulation tools (e.g., ns-3, ns3-gym) is desirable.
• Strong analytical skills and interest in decentralised AI, semantic communications, and IIoT systems design.
• Excellent communication skills in English; German is a plus.

Our offer:

• An exciting and innovative open research environment within a nationally funded R&D project.
• Direct collaboration with Shandong University (China) and academic research groups within the bilateral DFG/NSFC SemComIIoT project.
• International, dynamic work environment with highly qualified and motivated colleagues.
• Opportunity to publish research results and contribute to open tools, datasets, and a book on Semantic Communications for IIoT within the SemComIIoT project.
• Note: this master thesis position is not remunerated.

Have we piqued your interest?

If so, please submit your application, including a cover letter, a detailed resume, and a recent transcript.

Job ID: IIoT-MSC-03-2026
Contact: Prof. Dr. Rute Sofia personal@fortiss.org