Federated Learning for IoUT: Concepts, Applications, Challenges and Opportunities

I. INTRODUCTION

• IoUT links smart underwater devices and supports environmental monitoring, disaster prediction, exploration, shipwreck searches, tsunami prediction, and marine-life monitoring.The introduction describes autonomous underwater vehicles collecting and exchanging information with land-based control centers for predictive analytics.Source: §I · p. 1
• Underwater sensing can be damaged or interfered with, communication is difficult in variable conditions, and erroneous or irrelevant data can delay transmission and waste communication resources.The introduction links these conditions to inaccurate or insufficient data and challenges for mission-critical systems.Source: §I · pp. 1–2
• The introduction says machine learning can improve IoUT efficiency, fault tolerance, reliability, latency, information processing, and sensing compared with deterministic traditional technologies.The technologies discussed include acoustic links, RFID tags, and microbial fuel cells.Source: §I · p. 2

Role of Federated Learning in IoUT Systems

• FL keeps sub-model training at individual nodes and exchanges encrypted model parameters with a fusion node, aiming to reduce communication overhead while supporting security in IoUT.The introduction also describes edge nodes as processing data near its source to reduce useless data and transmission size.Source: §I · p. 2

I. INTRODUCTION

• The introduction presents FL for IoUT as motivated by a gap: the authors report no prior survey specifically on FL for IoUT, while earlier surveys covered FL in industrial and aerial settings and IoUT networking or marine-data analytics.The authors qualify the gap with “to the best of our knowledge.”Source: §I · p. 2
• The paper is organized to cover IoUT and FL techniques, FL applications in IoUT, integration challenges and future directions, and a conclusion.Source: §I · p. 2

II. BACKGROUND

• The section introduces IoUT.Source: §II · p. 2
• The section introduces FL.Source: §II · p. 2
• The section discusses the motivation for integrating FL with IoUT.Source: §II · p. 2

A. Internet of Underwater Things

• IoUT data supports ocean-resource management, shipwreck and crash surveys, early disaster detection, marine-life monitoring, environmental surveillance, and defense.Applications range from small observatories and harbors to ocean-trade coverage at global scale.Source: A. Internet of Underwater Things · p. 2
• Harsh conditions, depleted batteries, signal errors, interference, malicious attacks, and obstacles make reliable sensing difficult; AI/ML can model failure relationships and fuse diverse sensor data.Deep learning can analyze historical data to generate inferences and predictions in complex underwater environments.Source: A. Internet of Underwater Things · pp. 2–3
• Table I is presented as a summary of state-of-the-art implementations in IoUT.Source: Table I · p. 3
• Compromised sensor data and energy-limited underwater networks can leave nodes vulnerable to attacks that expose or manipulate data and contribute to inaccurate AI decisions.Source: A. Internet of Underwater Things · p. 3
• Underwater transmission faces bandwidth limits, Doppler effects, multipath interference, and acoustic-channel delays; multimodal communication and reinforcement learning are described as approaches to transmission challenges.The section names acoustic, radio, optical, and magnetic-inductive communication; it also notes routing and denial-of-service attacks.Source: A. Internet of Underwater Things · p. 3
• Centralized conventional ML requires sensor data to be sent to the cloud, creating communication costs and exposing sensitive data during transmission or storage.Source: A. Internet of Underwater Things · p. 3
• Rapidly generated IoUT data is difficult to label in real time and handle at scale, while network loss and external noise can reduce training-data quality.Source: A. Internet of Underwater Things · p. 3

B. Federated Learning

• FL distributes the global machine-learning model across individual devices.Source: B. Federated Learning · p. 3
• Each device executes the learning algorithm locally and transfers model updates, rather than raw data, for global training.Source: B. Federated Learning · p. 3
• Because raw data is not sent to the central server, the passage says local data is not exposed to potential malicious users or devices.Source: B. Federated Learning · p. 3

C. Motivation for Integration of Federated Learning for Internet of Underwater Things

• FL is motivated as a way to preserve sensitive data generated by IoUT devices.The motivation contrasts local computation with conventional approaches that transfer data.Source: C. Motivation for Integration of… · pp. 3–4
• FL is expected to reduce latency and communication costs during data transmission.Source: C. Motivation for Integration of… · p. 4
• The paper states that FL can handle noisy data better than traditional ML algorithms.Source: C. Motivation for Integration of… · p. 4
• FL is also motivated as a way to reduce data-quality issues during transmission of large IoUT datasets.Source: C. Motivation for Integration of… · p. 4

• The paper discusses applications of FL in different IoUT verticals.Source: §III · p. 4

A. Environmental Monitoring

• For water quality and oil-spill monitoring, sensors and drones send measurements to servers or sink nodes for timely decisions; one approach clusters monitoring stations for local training and server consolidation of model weights.Source: A. Environmental Monitoring · p. 4
• In the proposed model, encoding decreased by 9% for 20 MB of files, and more blocks could be transferred per second.Source: A. Environmental Monitoring · p. 4

B. Underwater Exploration

• For underwater exploration, FL locally trains on collected data rather than sharing it with a centralized server, supporting data, location, and device privacy.Source: B. Underwater Exploration · p. 4
• 97% accuracy was reported for ARC/FML on the DeepSink and RF channel datasets.Source: B. Underwater Exploration · p. 4
• At attack intensities of 20% and 80%, the blockchain-FL system achieved loss functions of 0.092 and 0.0195, respectively.Source: B. Underwater Exploration · p. 4

C. Disaster Prevention and Mitigation

• FL-enabled IoUT can collect continuous environmental data and learn shared disaster-prediction models across devices rather than storing the model centrally.The paper also identifies improved model performance and scalability as benefits for disaster prevention.Source: C. Disaster Prevention and Mitigation · p. 5

D. Defence/ Military

• UWSNs detect and classify underwater subjects; FL is proposed for MDO data-sharing constraints, where the US Navy seeks real-time sharing and Cirincione and Verma propose FL.Source: D. Defence/ Military · p. 5

E. Navigation and Localization

• Poor underwater positioning and navigation are predominant IoUT issues, while localization supports source detection and tracking.Sensor locations also serve as reference points for smart objects, swimmers, divers, and explorers.Source: E. Navigation and Localization · p. 5
• A continuous federated learning process is proposed for localization and navigation because the input is streaming data and decisions occur in seconds or milliseconds.Federated learning is described as collaborative learning that avoids restrictions on sharing training data.Source: E. Navigation and Localization · p. 5
• The paper identifies navigation and localization among IoUT application areas, with application verticals depicted in Fig. 3.Source: E. Navigation and Localization · p. 5, Figure 3 · p. 5

Device/ Network Configuration

• Mobile nodes, path loss, noise, multipath, and Doppler shift disrupt connectivity, while edge devices have limited battery, storage, and computation.The paper calls for lightweight models, self/auto-configuration, and hardware-algorithm co-design.Source: §IV · pp. 5–6

Data Transmission

• Underwater transmission interference and connectivity problems complicate synchronous FL, which waits for the slowest client before server aggregation.Asynchronous server updates are suggested as a research direction.Source: §IV · p. 6

Unreliable Channel Condition and System Heterogeneity

• Unreliable channels and changing network topology impede client updates, while heterogeneous devices and data formats complicate system design.The paper suggests optimization, gradient compression, efficient aggregation, and load distribution across devices according to availability.Source: §IV · p. 6

Privacy

• FL does not eliminate privacy risks: gradient information can enable user-data reconstruction, and model inversion can reconstruct collected images.Secure, lightweight protocols are suggested as a possible response.Source: §IV · p. 6

Real-time generation of labels

• Real-time class-label generation is time-consuming and tedious for supervised IoUT learning.Suggested options are unsupervised learning, which does not require labels, or automated real-time labeling tools.Source: §IV · p. 6

IV. CHALLENGES, OPEN ISSUES AND FUTURE DIRECTIONS

• Table II summarizes challenges and possible solutions for deploying FL in IoUT.Source: §IV · p. 6

V. CONCLUSION

• The paper reviews IoUT systems, their applications, and future research directions.Source: §V · p. 6
• It surveys FL integration in environmental monitoring, defence, underwater exploration, and disaster prevention.Source: §V · p. 6
• The review highlights potential challenges of deploying FL underwater, possible solutions, and scope for future research.Source: §V · p. 6