Federated learning (FL) allows multiple clients to train a machine learning model on a server without sharing their private data. To reach a consensus, the server collects alternative information such as model updates. The sub-field of heterogeneous FL investigates scenarios where clients have varying charac-teristics. This thesis investigates the model-heterogeneity challenge, where the clients train different model architectures, leading to compatibility issues in reaching a consensus. This thesis addresses model heterogeneity in a setting where each client owns multiple devices. We also assume that devices belonging to the same client can share data among themselves. This assumption is atypical in FL and prompts us to re-assess the limitations of knowledge transfer techniques such as knowledge distillation. We propose a set of tools centered around knowledge distillation that leverage the data-sharing policy to address model heterogeneity. We also quantify the levels of data and device heterogeneity present on the devices to
develop an adaptive solution that fits the type of heterogeneity encountered. We demonstrate the effectiveness of our solution in training a machine learning model in a federated setting with model heterogeneity by comparing it with alternative solutions in terms of latency and accuracy. Finally, we provide dir-ections for further research, such as devising tests for data heterogeneity and exploring other model compression techniques in heterogeneous FL.

Unmanned aerial vehicles (UAVs), often referred to as autonomous drones, are becoming more and more prevalent in our daily lives. Drones are usually equipped with traditional frame-based cameras and have functions like object detection. However, the high energy consumption of frame-based cameras presents a challenge to drone endurance. In addition, due to its own hardware limitations, problems such as blurring and deformation will occur when capturing high-speed moving objects. In contrast, event cameras, as one of the latest neural technology cameras, have the characteristics of low power consumption, low latency, and sensitivity to high-speed objects. This makes them well-suited for integration into drone platforms. In this thesis, we introduce a method for high-speed moving object recognition based on event cameras. This approach involves enhancing objects with externally added ``common features": fiducial markers, and employing a custom-developed deep learning neural network to detect these markers. These marks can carry some messages, such as the information of related objects. When these marks are detected, it is considered that objects are detected. We also developed a fiducial marker decoding method based on region segmentation to obtain the message content, thereby achieving interaction with the detected object. Evaluation results show that the proposed method has a low computing time of 26.9 ms, low storage and memory usages of 670 MB CPU memory and 750 MB GPU video memory, and a high accuracy of 77.9%, making it suitable for high-speed object recognition based on event cameras.

The increasing expansion of the electrical network accentuates the need for a better understanding of the quality of the existing infrastructure. Assessing the quality of the individual cables becomes instrumental in prioritizing replacements and grid reinforcements within the network. A critical evaluation of insulation quality in paper insulated lead covered cables is important to ensure a robust electricity grid. Moisture plays a key role in the degradation of the paper insulation. In this thesis a novel microcontroller based near-infrared spectroscopy system design is proposed which is able to measure the moisture content of paper insulated lead covered cable insulation with a theoretical accuracy of about ±1% moisture content, while being cost effective and portable. It achieves this accuracy by making use of an outlier detection algorithm which is able to increase the accuracy of the system from ±3% to ±1%, by removing outliers of individual sensors and averaging the results of multiple measurements. The resulting output of the algorithm is linked to the asset number of the paper insulated lead covered cable under test, and uploaded to the cloud to enable further data analysis. The proposed near-infrared spectroscopy system performs measurements on one particular absorption band of water; the absorption band around 1450 nm. Due to the influences of stray light, a design of a paper insulation sample holder is proposed. This sample holder is customized to the used components in the design and fixes the samples in place, which limits measurement inconsistencies. A testing procedure is proposed to link the sensor output to the moisture content of the paper insulation sample. A linear relation is found between the sensor output and the moisture content of the paper insulation samples at a moisture content of 0% to 20%.

Traditional search engines rely on centralized databases and powerful servers to process and retrieve information. Developing alternatives to key-value search engine databases in distributed computing environments is a significant challenge, particularly when dealing with limited computational resources. This study explores the use of large language models (LLMs) to address this problem. We focus on environments with constrained computing power, such as mobile devices, to investigate the feasibility of using LLMs as a localized search solution. Through experiments with the state-of-the-art LLMs BERT and T5, we demonstrate their ability to memorize and retrieve unstructured data, specifically YouTube video IDs, based on partial information derived from video titles or tags. Our results show that the explored models can achieve 100\% precision and recall when retrieving 48266 video IDs. The findings suggest that LLMs have the potential to effectively function as a search engine database, offering semantic search capabilities while operating within the constraints of limited computational resources.

There are increasing applications of Visible Light Communication, and LED-to-Camera communication is one of the promising research directions. In this paper, we explore LED-to-Camera communication in mobile scenarios based on the rolling shutter effect to get higher data transmission rate. Movement causes the deformation of the overall shape of the LED in the captured image. And stripes close to the LED edges are too dark to be detected. Both affect the data detection by utilizing the rolling shutter effect. A method is proposed that can increase the transmission rate, which solves the problems caused by movement and darkness of stripes that close to the LED cover edges. After processing, the region of interest is recouped, i.e., the processing range used to detect the rolling shutter stripes becomes larger, which contains more complete data. An approach to detect stripe edges by converting pixel information into binary signal is proposed, which is used to solve the problem of undetectable stripe edges due to high frequency. The proposed method is evaluated through experiments under different conditions. The results demonstrate that the method is effective for addressing the challenges of LED-to-camera communication in mobile scenarios.

With the rapid development of Artificial Intelligence (AI), the size and complexity of models are increasing rapidly. The limited memory and computing power of microcontroller units (MCUs) pose significant challenges for running AI applications on them. This thesis presents a method to run deep learning models on MCUs using a distributed approach.

First, we identified memory size as the primary constraint for implementing deep learning models in MCUs. To address this issue, our method involves splitting the models into smaller weight fragments distributed across multiple networked MCUs. A coordinator MCU manages the overall process, including neuron mapping and data relaying. We demonstrated that our approach reduces peak RAM usage during inference compared to existing methods.

For optimization, we employed layer fusion and quantization to reduce model size while preserving accuracy. We also introduced a rating system to assign capability scores to MCUs for efficient task allocation, explained through mathematical equations.

Our simulation phase validated the effectiveness of our method, showing successful distributed inference in MCUs and providing insights for real-world applications. Implementation on a network of MCUs confirmed the practical applicability and efficiency of our approach.

In conclusion, this thesis presents a feasible and efficient distributed inference method for networked MCUs, addressing resource limitations and enabling practical AI applications on constrained MCU platforms.

In order to have a scalable battery system that can store a lot of charge and output a large amount of power, it is possible to connect multiple battery packs in parallel. This does not come without its own caveats though. During a charge and discharge cycle of a battery pack, the voltage that is present at the output varies. In a scenario where two or more battery packs are connected but have different levels of charge, it is possible that the voltage difference between them is so high that the control system will not allow them to discharge together due to the possibility of large currents flowing between the batteries, which can lead to short and/ or long term damage to the battery. This presents a problem since by only using a percentage of the packs you have connected in parallel, you don’t get the maximum power output or capacity that you could have. The solution to this is to introduce some balancing system. This system will act- ively move charge from one pack to another and allow the voltages to equalise meaning the packs can discharge into a load together again. In this thesis paper, a method for performing this balancing is described, modelled, implemented and verified. It was found that for this specific imple- mentation the time it takes to balance a full battery pack with an empty battery pack is 13 hours and 39 minutes. This full-balance was done by using a PWM signal to limit the current between two packs. Another method was also im- plemented making use of parasitic characteristics of a MOSFET. However, this only served as a method to balance when the battery packs voltages are already somewhat close. Lastly, this thesis provides several fronts for improving the balancing systems and how the implementation can be further fleshed out to shorten the time it takes to balance two packs and how to make it more robust.

Spectrum sensing is a vital technology for alleviating pressure on the radio spectrum and will become more sophisticated as billions more devices come online. In the future, more advanced techniques utilizing deep learning will sense which parts of the spectrum are available to communicate on. This is a heavily researched area, but few papers demonstrate methods of deploying deep learning on the resource-constrained edge devices that will ultimately use them. One approach called Spectrum Painting augments spectrograms and detects which signals are present with a Convolutional Neural Network. We optimize this method for microcontrollers by simplifying the computation of spectrograms and using TinyML techniques. This results in over 90% accuracy on signals with a high signal-to-noise ratio and a latency of 159 ms on a 64 MHz CPU. Our findings conclude microcontrollers are capable of utilizing deep learning for spectrum sensing, but custom hardware will still be required in a real-world deployment.

Nowadays, the popularity of machine learning and artificial intelligence algorithms is very high. A new research direction has emerged where the machine learning algorithms are executed on resource-constrained embedded devices. With the development of the Internet of Things paradigm, these edge devices are deployed in a lot of places. Due to the limited resources of embedded devices, it is difficult to bring machine learning algorithms to them. This is where the on-device split inference comes in. It is possible to distribute the inference between multiple edge devices and the cloud so that the edge devices can execute the inference of complex machine learning models. This paper presents a systematized literature review of papers that focus on on-device split inference for edge devices. The papers are analyzed and compared based on pre-determined questions and displayed based on several categories.

Visible light positioning (VLP) systems enable indoor positioning through a deployment of light-emitting diodes (LEDs) as transmitters and photodiodes (PDs) as receivers. A promising approach in VLP involves recording the received signal strength (RSS) to construct fingerprint samples for later use in positioning. However, achieving high accuracy demands a labor-intensive data collection process.

In this study, we propose improvements to a data cleaning and augmentation pipeline. Our improvements focus on preserving more source data during cleaning and data-based LED position estimation for more reliable data augmentation. Experimental results show that our approach maintains comparable positioning accuracy while reducing data collection efforts by over 99%. Furthermore, we conduct experiments to investigate the impact of spatially irregular data collection strategies on positioning accuracy. Finally, we deploy a machine learning model on a microcontroller to demonstrate the practical feasibility of our proposed methods.

This paper, in answering the question ”Can effi- cient on-device spectrum sensing be achieved on microcontrollers?”, presents a simple yet compre- hensive approach to signal classification using Con- volutional Neural Networks (CNNs) optimized for deployment on resource-constrained devices. Us- ing data generated via MATLAB’s Wireless Tool- box, as well real world data obtained from testbeds, we created a robust dataset of 9000 samples for training our model. The steps we took while de- veloping a CNN model that performs efficiently on microcontrollers include: data augmentation (pre- processing), model compression and quantization. The model significantly outperformed baseline ac- curacy metrics and maintained competitive infer- ence times, despite the hardware limitations of mi- crocontrollers. This reinforces the idea that Deep Learning has great potential in signal classification. Our research has the potential of being applied to smart homes, IoT networks, industrial automation, and public safety, where our optimized model facil- itates efficient spectrum utilization and minimizes interference.

The explosive growth in data collection driven by the proliferation of interconnected devices necessitates novel approaches to data processing. Traditional centralised data processing methods are increasingly inadequate due to the sheer volume of data generated. Distributed Tiny Learning (DTL) offers a compelling solution by distributing machine learning tasks across multiple edge devices and processing data locally, thus minimising the need for data transmission to central servers. This approach is particularly beneficial in scenarios with limited network bandwidth and stringent privacy requirements, enhancing data security and compliance with privacy regulations. The advent of 6G networks, with their promise of unprecedented speed, capacity, and reliability, can further amplify the power of DTL. By providing higher bandwidth and lower latency, 6G enables more efficient data processing and communication among edge devices, thereby enhancing the overall performance and scalability of DTL systems. This integration supports real-time decision-making for applications such as autonomous vehicles, smart cities, and healthcare monitoring.

This paper conducts a comprehensive survey of the state-of-the-art in DTL, categorising scientific literature, mapping the ecosystem and tools, and addressing performance, efficiency, and scalability challenges in ultra-low-power devices within a 6G context. Additionally, it implements and benchmarks two DTL algorithms, providing practical insights into their effectiveness and operational viability.

Occupancy count is a critical metric for efficient smart building management. Despite advancements in sensor technologies, achieving accurate and robust low-cost privacy-safe occupancy counting remains to this day an unsolved challenge. Visible Light Sensing has in recent advances been shown to be a potential viable solution for such a system. This thesis aims to test the viability of VLS for occupancy counting, more specifically through the use of Spectral Light Information. Spectral Light Information is obtained via a single color sensor that decomposes ambient light into a finite number of channels. In this work, we use an off-the-shelf TCS34725. This thesis documents the development of this SpectraCount system, along with its analysis. To try and reach a robust and accurate final system, we propose to integrate our SLI-based system with a state-of-the-art commercial solution. This solution is aligned with our plug-and-play, privacy, and cost requirements. It utilizes thermopile arrays, along with background removal and blob detection technology. We use use sensor fusion to try and correct the inherent errors of both solutions. Moreover, we compare the result of SpectraCount to the previous only thermopile-based solution. The novel algorithm for color-sensor-based occupancy detection and count achieves an accuracy of 79.67\% as a standalone system for detection and 52.8\% for counting. By exploiting the advantages of both thermal and visible light sensing technologies, we obtain an improvement of 55.4\% compared to only thermal solutions, and no improvement over the standalone SpectraCount system.

This research investigates the impact of multipath signals in UWB communications and explores their potential to improve localization accuracy of tags using the additional information captured in the Channel Impulse Response (CIR). While traditional localization typically relies on multiple anchors, this study focuses on using a single anchor and creating multiple Virtual Anchors (VAs) through MultiPath Components (MPCs). The research proposes algorithms such as likelihood, algebraic, and fminsearch methods, and tests their performance through simulations in environments with anchors, tags, and reflective surfaces.

After confirming the feasibility of using reflection multipath signals for localization through the simulations, experiments were conducted with NXP's SR150 UWB tags, which collect CIR data via two antennas. An indoor environment was set up with two MPCs and a strategically placed anchor. Data was collected from various tag locations and analyzed using the different localization methods.

The collected CIR data revealed additional reflections, notably from the ceiling and ground, which sometimes interfered with the desired MPC reflections. To address this, data cleaning techniques were implemented for the likelihood method, and the two antennas were used to approximate the AoA in the fminsearch and multilateration methods.

The final results demonstrated that the likelihood method achieved an average localization error of 26 cm, the fminsearch method 20.7 cm, and the algebraic method 24 cm. While the fminsearch method provided the best accuracy, it required data from two antennas, unlike the likelihood method, which only needed one. The fminsearch method showed improved accuracy near MPCs, though its performance declined as the tag moved further away. This suggests its potential for applications where AoA is large, as traditional methods become less accurate beyond 60-degree angles.

Large language models (LMs) are increasingly used in critical tasks, making it important that these models can be trusted. The confidence an LM assigns to its prediction is often used to indicate how much trust can be placed in that prediction. However, a high confidence can be incorrectly trusted if it turns out to be incorrect, also known as a high-confident error, or unknown unknown (UU). Blind spots, clusters of UUs, are caused by out-of-distribution data (OOD), bias, covariate shift, or unseen data. Previous work by Lippmann et al. generated samples based on these UUs, however, they selected a random subset of all found UUs. Although this was the first study that could mitigate blind spots without the need for crowd workers or oracles, they did not address the underlying causes of blind spots, such as OOD and bias. Data valuation is the research field that is concerned with valuing samples to improve model performance and can identify OOD data, bias, and noise. This paper proposed to combine the research field of data valuation, with the mitigation of blind spots. We generated synthetic samples using the highest-valued samples and retrained the LM using a Weighted Loss based on Data Values (WLDV). We conducted an extensive evaluation of our approach on four tasks, demonstrating a reduction of UUs by up to 32.7%, while retaining the same level of accuracy. This was the first exploration of combining the research fields of data valuation and blind spot mitigation.

This thesis presents LightLetter, a system that is designed for recognizing fingertip air-writing of both numbers and letters. This provides a cost-effective and privacy-conscious method for interacting with public devices, such as touchscreens. The current iteration of LightLetter allows for the input of letters and numbers to public devices using a 3×3 photodiode array, leveraging the ambient light in the environment.

The system’s success hinges on a method that adeptly processes raw data from the photodiodes, transforming it into a standardized trajectory image. To extract these trajectories, we employed the optical flow extraction method and then used two separate networks for their transformation. Initially, both the direction and magnitude of finger movements are derived from the raw data with the assistance of the Farnebäck optical flow method. Subsequently, the principal component of these trajectories is preserved, representing the finger’s path during that interval.

For the purpose of generating standardized trajectory images, the CUT model is trained using both the extracted trajectories and the EMNIST dataset, which serves as the unpaired corresponding dataset. The generated images, when paired with the extracted trajectories, facilitate the training of the pix2pix model which is served for the inference phase. In the inference phase, the Farnebäck optical flow method and the pix2pix model are combined to batch-produce standardized trajectories. These are then recognized by a model that is trained using the EMNIST dataset.

Our results of 79% accuracy on alphanumeric data and 99% accuracy on digits indicate a marked enhancement in the recognition capabilities of the model trained with this method compared to the baseline data which has 90% accuracy on digit recognition. Moreover, the model demonstrates superior generalization and robust recognition prowess.

In recent years, there has been a growing interest among researchers in the explainability, fairness, and robustness of Computer Vision models. While studies have explored the usability of these models for end users, limited research has delved into the challenges and requirements faced by researchers investigating these requirements. This study addresses this gap through a mixed-method approach, involving 20 semi-structured interviews with researchers and a comprehensive literature analysis.
Through this investigation, we have identified a practical need for a data model that encompasses the essential information researchers require to enhance explainability, fairness, and robustness in Computer Vision applications. We developed a data model that holds the potential to improve transparency and reproducibility within this field, speed up the research process, and facilitate comprehensive evaluations, whether quantitative or qualitative, of proposed methodologies. To refine and demonstrate the practicality of the data model, we have populated it with four existing datasets. Additionally, we have conducted two user studies to validate the model's usability. We found that participants were enthusiastic about using the data model. Some potential uses described by the participants were comparing models and datasets, accessing (niche) datasets and models, creating and exploring datasets, and having access to ground truth explanations. However, participants also had concerns about the data model, mainly with its usability being restricted to people with database knowledge and the richness of data in the database. Nonetheless, hope that this research constitutes the first step for data modelling for researchers in the field of Trustworthy AI.

With the development of 5G/6G communication technology, an increasing number of communication protocols for data transmission are created and implemented. At the same time, the redundancy of packet headers has become a matter to be optimized when transferring packets across different protocol layers, which causes high resource occupancy, low transmission efficiency and high latency, especially for real-time communications like Voice over IP (VoIP).

The currently standardized Robust Header Compression (RoHC) Algorithm seeks to provide an approach to compressing the protocol headers inscribed in data packets to decrease the amount of data transmission without reducing the communication quality. However, the existing RoHC algorithm is not efficient enough for the requirements of present header compression, as it's initially designed for specific headers like RTP, UDP, and IP. The packet headers must be optimized over the protocol stack for the real-time data stream and the GPRS Tunneling Protocol (GTP) tunnel sections in the 5G network.

An Optimized Header (OH) Compression Algorithm based on the existing RoHC algorithm is proposed in this thesis to compress the headers over the 5G protocol stack in a coordinated fashion for real-time transmission and GTP tunnels. Besides, further development of the OH algorithm is done to optimize protocol headers by merging duplicate header fields for both multiple QoS flows and multi-layer Protocol Data Unit (PDU) flows in the same PDU session. The overall optimization is conducted in the 5G System (5GS) over F1-U and N3 reference points.

The simulation results indicate that the OH algorithm improves the average header compression rate by 45% compared with the RoHC algorithm, as demonstrated via scenario-based 5GS simulations. The average number of transported bits is decreased by 30%, 15% and 26% for Opus coding audio stream, H.264-Opus coding video-audio stream and BLE Version 4.2 triple packets for smartwatch telemedicine stream, respectively, compared with the RoHC algorithm. Besides, the average algorithm execution latency is decreased by 2 μs compared with the RoHC algorithm. Moreover, the multi-stream and multi-layer optimization for the OH algorithm brings about a higher header compression rate and lower execution time in specific scenarios. The overall results indicate that the OH algorithm is capable of compressing the packet size effectively, decreasing the number of transported bits, and shortening the execution time, resulting in higher transmission efficiency and lower latency for VoIP services.

This thesis presents Screen Antenna - A Visible Light Communication (VLC) system that integrates data transmission and reception, with the conventional pixel display capability of RGB LEDs. The system is constructed with off-the-shelf components and runs on the Arduino Due microcontroller. The hardware and software have been designed with the objectives of scalability and flexibility. This research aims to optimize the system's performance in terms of throughput, error rate, and display quality.

To evaluate the system's performance, extensive experiments were conducted under different lighting conditions, PWM frequency and Rx-Tx distances. The experiments focused on achieving the maximum achievable throughput while maintaining a low Bit Error Rate (BER). Through careful optimization of the system parameters, a maximum throughput of 4.4kbps was achieved. Furthermore, the system consistently maintained a Bit Error Rate of less than 0.005\%, ensuring reliable and error-free data transmission.
In addition to data transmission, the impact of data transmission on the display's visual quality was investigated. The Delta E value, which quantifies the observed colour difference between two pixel values, was used as a metric for display quality. The data rates were calibrated such that the display operated with a Delta-E value of less than 1 in more than 90\% of the colour spectrum, indicating minimal colour distortion during data transmission.

This study contributes to the field of VLC by showcasing a practical implementation that combines scalability, visual quality, and performance. Future research directions could focus on expanding the display size and enhancing the data rate by means of more sophisticated hardware.

Visible light sensing is a field of research that creates new possibilities for human-computer interaction. This research shows the viability of designing a system for detecting hand gestures using a cost-effective detection circuit employing 3 light-sensitive photodiodes. The way this research shows viability is by developing a machine-learning model that works on 3D-structured sensor data that is able to distinguish 10 different gestures and deploying the model on a standalone Arduino Nano 33 BLE microcontroller controlling the system. Using a combination of Convolutional Neural Networks and Recurrent Neural Networks it is possible to deploy a model called ConvLSTM-128 that achieves an accuracy of 70% on a dataset of limited size. This research acknowledges that the achieved accuracy is not suitable for real-world use, but concludes by outlining steps that could help future research in increasing the accuracy. Furthermore, an analysis of the 10 gestures shows that in order to improve accuracy, the way some gestures are performed might need alteration. Finally, a model size of around 140Kb and an inference time of 660ms show that this model is compact and fast enough to be deployed in real-world applications.