We present a decentralised alternative to the winner-takes-everything dynamics of social media platforms.

For 25 years there have been continuous attempts to decentralise file sharing, music streaming, video conferencing, and social media. 

None of these hundreds of projects to re-decentralise the Internet have reached the uptake level of YouTube and TikTok. They are rarely easy to use.

We present DeToks, a fully decentralised alternative to Youtube and Tiktok. DeToks is not dependant on any central server or cloud. DeToks is specifically designed to be as decentralised and attack-resilient as Bitcoin and Bittorrent.

Our core contribution is effortless 5G Network Address Translators (NATs) puncturing.

Direct phone-to-phone communication is not available on today's smartphones. DeToks solves this problem.

NATs and carrier-grade NATs block direct communication between smartphones. 

We procured 30+ SIMs card on European 4G/5G mobile networks and measured the carrier-grade NATs behavior. We determined the NAT types (full cone,restricted,symmetric) and their time-out settings. By leveraging provider-aware (Vodafone,Orange,Telia, etc.) NAT puncturing strategies we create direct UDP-based phone-to-phone connectivity. 

We utilise parallelism by opening at least 500 Internet datagram sockets on two devices. By relying on provider-aware IPv4 range allocations, provider-aware port prediction heuristics, high bandwidth probing, and the birthday paradox we can successfully bypass even symmetric NATs. Our communication method achieves peer-to-peer 5G connectivity at the cost of merely some initial delay and bandwidth, without any assistance from third party servers or clouds.

Detoks validates our 5G puncturing work. We demonstrate the feasibility of fully decentralized social media platforms on consumer mobile devices.

The demand for embedded devices, such as intelligent sensors, smartwatches, medical implants, and computing chips in cars, has been rising steadily in the past and is expected to continue for the coming decade. Their high and increasing degree of connectivity expands the attack surface for attackers, which is exaggerated by their physical accessibility. This makes them particularly vulnerable to invasive attacks. Compounding these risks is the growing reliance on third-party software. An example of such risks can be seen in the recent incident where a software solution from the company CrowdStrike, integrated into the Windows Operating System (OS), caused a major global outage that left many airports, hospitals, banks, and government departments worldwide unable to operate. The Guardian reports that this incident affected 8.5 million Windows machines and resulted in an estimated financial loss of 5.4 billion USD for companies in the US alone. Similar risks exist for OSs for embedded devices. These devices do not run general-purpose OSs like Windows, however, in their very nature they are similar. They provide a runtime abstraction for user processes and manage resources for which the OS requires higher privileges. FreeRTOS is a popular embedded OS for resource-constrained devices, also offering real-time capabilities, making it a Real-Time Operating System (RTOS). Meanwhile, RISC-V, a relatively new Instruction Set Architecture (ISA), is gaining popularity due to its open-source philosophy. However, despite the increasing popularity of RISC-V and the widespread use of FreeRTOS, there is a lack of comprehensive security support for FreeRTOS on RISC-V platforms.

This thesis addresses the lack of security for malicious third-party applications and presents a novel FreeRTOS implementation method on a RISC-V embedded system platform. The security of FreeRTOS is enhanced by leveraging the Physical Memory Protection (PMP) feature of RISC-V. The primary contribution involves the FreeRTOS kernel dynamically manage the access rights using PMP, preventing malicious tasks from causing harm. This dramatically improves the security, because the flat memory structure inherent to FreeRTOS is prone to being readily exploited and compromised by attackers. Further, an overview of known attacks on embedded systems and countermeasures is presented, a subset of which are explored in more detail. A secure platform is proposed that aims to demonstrate a holistic approach to integrate the FreeRTOS with PMP support into a System-on-Chip (SoC). Then, the FreeRTOS platform is augmented by a secure boot procedure and a memory encryption unit that ensures the integrity and confidentiality of the external memory. This encryption unit uses the Prince encryption scheme and is elegantly combined with RISC-V’s PMP functionality. Unused bits in the PMP configuration registers are used to control the encryption selectively for single PMP memory regions. Moreover, an address-based encryption mode for this unit is proposed that addresses the weakness that plaintext data is always encrypted to the same ciphertext. This is problematic because an attacker could analyze the structure of the ciphertext and copy the ciphertext of memory cells with known plaintext. The effectiveness of the PMP task isolation is demonstrated by simulating malicious tasks that try to access specific data addresses in the memory of other tasks and the kernel. This setup imitates real-life attack scenarios, assuming the addresses have been exposed to the malicious task through reverse-engineering of the application code. The secure boot is validated by implementing the design on an FPGA and manually tampering the memory, resulting in a failed boot-up. Additionally, the impact of the PMP integration on the real-time capabilities of FreeRTOS is investigated more thoroughly by evaluating real-time metrics: task switch time, access time for shared resources using a mutex (a software construct to avoid access collisions), and dynamic memory allocation. The system design incorporates a RISC-V core named CV32E40S, an open-source core from OpenHW Group. All the security enhancements were run on hardware with a varying number of active PMP regions. In the worst-case scenario, using 64 PMP regions results in a 2.1x increase in task switch time, a 2.4x increase in the time to access shared resources using a mutex, and a 1.7x increase in memory allocation time. The area of the CV32E40S, measured in Look-Up-Tables (LUTs), increases with the number of regions: a 2.2x increase for 16 regions, 3.4x for 32 regions, and 5.4x for 64 regions, compared to the core configuration without PMP. Despite these increases, it remains within an acceptable range relative to the total area of the SoC including cache memory and peripherals.

This thesis presents the development and evaluation of a real-time neural network-based audio classification system designed on an NXP HW board to distinguish emergency response vehicles by their sirens from other vehicles. At the core of the system is a deep learning model that processes audio inputs captured via a microphone, classifying them based on the presence of siren sounds. The system achieves this by extracting audio features and running inference through the designed neural network, followed by post-processing to detect sirens accurately. Audio signals are transformed into mel-spectrograms, which represent the frequency spectrum over time using a specific window size for analysis. The neural network leverages these mel-spectrogram features to perform audio classification.
The deployment of this system involves several key steps. First, the model is trained on diverse audio data, including siren and non-siren sounds. Audio signals are transformed into mel-spectrograms, which capture the frequency spectrum over time. The neural network processes these features to classify the audio based on the presence of siren sounds. The classified results undergo post-processing to enhance detection accuracy. The system is tested in real-world scenarios, demonstrating a turnaround time of less than 3 seconds even under high noise conditions. Various trade-offs are evaluated to improve efficiency, reduce memory size, and minimize latency, ensuring the system meets requirements for model size, latency, and compute cycles. The custom dataset comprises 280 hours of audio, including well-known, publicly available datasets such as ESC-50, Audioset, and UrbanSound. This dataset is enriched with both original and augmented siren sounds and non-siren audio to enhance the model’s learning efficacy and robustness. The system achieves a 96.19% test accuracy in identifying sirens and is suitable for deployment in real-world scenarios, even for SNRs as high as -30 dB. Although the system meets most requirements for model size, latency, and compute cycles, the false positive rate needs improvement. This can be achieved by expanding the dataset and retraining the model.

Deep learning models are now widely deployed on edge IoT devices. However, most of these models are trained under supervised conditions and can only recognize seen classes learned from the training stage. Zero-shot learning (ZSL) is a popular method for identifying unseen classes by leveraging the semantic information from both seen and unseen classes. Foundation models (FMs) trained on web-scale data have shown impressive ZSL capability in natural language processing and visual understanding. However, leveraging FMs' generalized knowledge for zero-shot Internet of Things (IoT) sensing using signals such as mmWave, IMU, and Wi-Fi has not been fully investigated. In this work, we align the IoT data embeddings with the semantic embeddings generated by an FM's text encoder for zero-shot IoT sensing. To utilize the physics principles governing the generation of IoT sensor signals to derive more effective prompts for semantic embedding extraction, we propose to use a multi-source information fusion strategy, cross-attention, to combine a hard prompt generated by Large Language Models (LLMs) and a soft prompt consisting of learnable vectors. To address the problem of IoT embeddings biasing to seen classes due to the lack of unseen class data during training, we propose using data augmentation to synthesize unseen class IoT data for fine-tuning the IoT feature extractor and embedding projector. We evaluate our approach on multiple IoT sensing tasks. Experiment results show that our approach achieves an average improvement of 1.0% in open-set detection and 9.5% in generalized zero-shot learning compared with multiple baselines on three datasets.

The classification of bugs in functional languages is an understudied area, as opposed to imperative counterparts, such as Java. This paper acts as an initial step to cover this gap into two complementary directions. First, a dataset of 142 bugs from 10 Haskell FOSS repositories have been classified according to two taxonomies from literature in order to assess how well they handle the differences in programming paradigms. Based on our bug classifications, the first taxonomy has very little variance in types of bugs, with 86% of bugs being classified in the same category. On the other hand, the second taxonomy showed more potential as the bugs were more balanced between categories, but with occasional difficulties in classification, as some bugs fitted more than one type. Second, we performed interviews with four Haskell developers about their experience with bugs and the usefulness of these taxonomies in practice. They argued that while such taxonomies can prove useful, the context in which it is being used is more important. Thus, circumstances such as the size of the project and team, and stages of development need to be taken in consideration before trying to apply any taxonomy.

This study explores the use of Code Churn and Pattern Size (PSIZ) metrics to identify bug-prone areas in Haskell codebases. The primary research questions addressed are whether these metrics can effectively predict areas of code instability and potential bugs. Our contributions include a comprehensive analysis of these metrics across three large Haskell projects, examining the correlation between high metric values and documented bugs. Our findings reveal that Code Churn is a significant indicator of potential bugs, with ’buggy’ files showing markedly higher mean code churn values. However, the PSIZ metric proved ineffective in predicting bug-prone areas, as the mean and median values for both projects and ’buggy’ files were similar and low. These results suggest that while Code Churn is a useful metric for identifying unstable code areas, PSIZ does not offer the same predictive value. Future research should expand the dataset and consider additional metrics to enhance the reliability of these findings.

Accurate topic classification is crucial in the scientific community when it comes to finding relevant journals. However, the efficiency and accuracy of topic classification of publications do not seem to be at its best performance, especially with the fast-paced rise in the quantity of research papers. Our research aims to address this problem by utilizing state-of-the-art (SOTA) methods. We chose the 'April 2022 Crossref' data set for the research, as Alexandria3k, the tool utilized for querying on the open data set, is tested on the same data. We stratified 50,000 data that have title, abstract, and work names, which are the roughly assigned topics. SOTA methods chosen for feature extraction and classification models are OpenAI Embeddings and XGBoost. Our research shows that this combination of SOTA methods has the potential to improve the performance of current topic classification of publications.

The growth of academic publications, heterogeneity of datasets and the absence of a globally accepted organization identifier introduce the challenge of affiliation disambiguation in bibliographic databases. In this paper, we create a baseline using the currently implemented algorithm for author affiliation linkage in Alexandria3k by comparing it to the ground truth. We aim to explore the usage of LLMs (GPT-4) in the Alexandria3k environment to disambiguate author affiliations. The proposed approach extracts the research organization from textual affiliations provided by researchers through their published works and cross-references the organization across the Research Organization Registry. Our process shows promising results and a significant improvement on the existing algorithm in terms of matching rate and identification of multiple affiliations. We discuss the margin of error in LLM results, limitations of the ground truth, and suggest future research directions.

The Alexandria3K library, a versatile Python-based tool, has been expanded to include the integra- tion of the PubMed dataset, enriching its capabil- ities in the analysis of scientific papers. Origi- nally supporting major datasets like Crossref and US patents, and smaller yet significant datasets like ORCID. The addition of PubMed enables in-depth analysis of medical papers, with medical specific data and articles not yet in the Crossref dataset. This research focused on validating the integration of PubMed into Alexandria3K. To achieve this, two literature surveys were replicated using the com- plete PubMed dataset. The first survey involved querying different pathogens in the dataset for three regions. The results were comparable, although some articles were missed by Alexandria3K but two articles were also missed by the original sur- vey. The second survey revolved around software tools used in medical papers. Although fewer ar- ticles were found with Alexandria3K the ratio for most tools was still comparable. Although a thor- ough manual review of all articles could have fur- ther refined the reevaluation, time constraints pre- vented this step. These replicated surveys demon- strate Alexandria3K’s potential in conducting lit- erature surveys, underscoring the need for manual validation to complement its capabilities.

This study addresses the gap for fine-grained emotion recognition in immersive environments utilizing solely data from on-board sensors. Two data representations of users eyes are utilized, including periocular recordings and eye movements (gaze estimation and pupil measurements). A novel multi-representation method integrating feature extractors for each representation alongside an effective feature fusion technique is proposed. The method significantly outperforms baselines that use only a single representation or incorporate content stimuli. It achieves an F1-score of 0.85 with 10\% data, approximately 40 seconds of data from all emotions, for personal adaptation, recognizing emotions while watching unseen parts of the stimuli used for adaptation. In a more practical scenario, the method achieves an F1-score of 0.71 with five seconds of personal adaptation data from each emotion, recognizing emotions while watching completely unseen stimuli. Under the same but more extreme condition, where only one second of data is available, the proposed achieves an F1-score of 0.68.
Furthermore, the study demonstrates that estimated labels can substitute for user-provided labels without sacrificing performance in emotion recognition, thus eliminating the need for users to manually label emotion elicitation segments.
Future work will focus on improving performance by allocating more computational resources and making architectural modifications, conducting deeper investigations into the decision-making process, and developing real-time recognition systems for in-the-wild experiments.
The results of this study suggest that more engaging, adaptive, and personalized experiences in immersive environments can be developed.

Author name disambiguation, otherwise described as (publication) record linking, is a problem that has had considerable research dedicated to its solv- ing. Author attributions, calculating research met- rics and conducting literature reviews are amongst processes that experience increased difficulty due to ambiguous author names. In this study, a novel approach is presented to disambiguate au- thors related to scientific publications, using Large Language Models (LLMs) in combination with the Alexandria3k software package. LLMs have shown great potential in processing, analysing and drawing conclusions when presented with human- readable data. The approach presented in this study supplies a LLM with known attributes of publica- tion records and authors, such as names, affiliations and co-authors, to determine whether records writ- ten by authors with ambiguous names can be linked to the same real-world person. Using Alexan- dria3k, a dataset of authors and publications with confirmed identities is created to test and validate the approach. Finally, the approach is measured against state-of-the-art methods to disambiguate author names and different configurations are pre- sented and discussed.

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.

Hidden spy cameras are a growing worldwide threat to people’s intimacy and privacy. With the growing interest in full-screen devices and the underlying development of under-screen cameras, a new type of potential security risk is introduced. Recent smartphones such as the ZTE AXON 40 already demonstrate that it’s infeasible to detect the camera with the human eye. There exist several techniques to detect hidden cameras, however most of these techniques are not resilient to the unique deployment scenario of the under-screen camera. A recent optical detection technique, which relies on the retro-reflective effect of hidden cameras, is promising but is also greatly hindered by challenges introduced due to reflections from the screen that is placed in front of the under-screen camera. In this work, these challenges are addressed, by proposing a detection principle that exploits the difference in reflective nature between the USC and the screen. Using reflection detection in a sliding window approach, a detection methodology is given to detect the USC. Furthermore, a detection architecture is designed that incorporates the proposed detection principles using a combination of computer vision, image processing and machine learning techniques. Using an off-the-shelf Time-of-Flight sensor, this architecture is implemented into a detection system and evaluated on its robustness and detection accuracy. Experiments on a dataset of 200 videos with a variety of measurement conditions show that this detection system is capable of achieving a USC detection rate of 71.5% while having a false-positive rate of 21.5%. It also proves excellent results while the screen is displaying content.

The behaviour of software is intrinsically linked to the hardware it runs on. As hardware innovations continuously change the landscape of computing, software has to adapt to these changes. Running legacy software on new hardware requires either the old hardware to be emulated, or a very time-consuming and error-prone process of reverse engineering the software to determine its behaviour, and then writing new software that has the same behaviour, but runs on the new hardware. Binary lifting and translation tools aim to automate this process, but are often limited by the lack of accurate and complete instruction semantics.

This project aims to evaluate the feasibility of using the Arm Machine Readable Specification to aid in this process. The specification is a machine-readable description of the Arm architecture, including self-proclaimed "executable" instruction semantics, written in a specification language called ASL. This project has created an emulator that can run Arm instructions on a non-Arm architecture, using the specification to determine the behaviour of the instructions. The emulator is able to run simple programs with little context, but is not yet able to run more complex programs, due to the lack of support for behaviour that depends on the specific hardware implementation, and context dependencies outside of the instruction set, such as operating system interactions.

This emulation based solely on the specification has demonstrated that it is a promising approach to reason about the behaviour of Arm programs, but it is not complete enough to be used for binary lifting and translation. The context dependency of executables, and the lack of implementation specific behaviour in the specification are the main reasons for this.

Refactoring is a useful tool for increasing the overall quality of software without making changes to how it interacts with the environment. To verify that a refactoring operation correctly transforms an expression, one can provide a formal proof. Using Agda, a dependently-typed language, as a proof assistant, we investigate the feasibility of proving the correctness of refactoring tuples to records for a small-scale language that shares similarities with Haskell. We construct this language in Agda using intrinsically-typed terms and define an accompanying refactoring function for refactoring tuples to records. We prove that the refactoring is well-typed and that it replaces all tuple occurrences. Big-step semantics are used to show the relation between the intrinsically-typed language and its resulting output value. Additionally, we show that we can construct a relation between the values of an expression before and after refactoring. By presenting these proofs we gain more insights into the feasibility of proving the correctness of tuple to record refactoring. Furthermore, we argue that the proofs given for this small-scale language can serve as inspiration for proving comparable properties of the refactoring in the context of Haskell and beyond.

This paper concerns itself with correct by construction refactoring of Maybe values to List values in a Haskell-like language (HLL) as a case study on data-oriented refactorings. Our language makes use of intrinsically-typed syntax and de Bruijn indices for variables. Operational semantics are defined using big step semantics. We define a refactoring function which intrinsically verifies its well-typedness due to our intrinsically-typed syntax. The semantic validity of this refactoring is given by a separate proof. We use Agda, a functional language and theorem prover, to define and prove these properties. Techniques and concepts used in our refactoring and proof generalize well to other data-oriented refactorings.

Refactoring tools are an important tool for developers, but their reliability can be questionable at times. In this paper, we show that it is feasible to formally verify refactoring tools using computer-aided proofs. To this end, we create a Haskell-like language and a refactoring operation on this language to add an extra function argument to an arbitrary function in the program. And finally, use the Agda proof assistant to construct a proof of the correctness of this refactoring.

When designing critical software, great care must be taken to guarantee its correctness. Refactoring is one of the techniques used to improve code readability, maintainability, and other factors without changing functionality. Thus, to ensure that it is properly applied, automated tools are used to perform refactoring. To ensure that the code remained correct, we verified the correctness of the tool actions using formal proof. This is assisted by a dependently typed language Agda, which, when used as a proof assistant, can directly support us by ensuring the soundness of the steps taken. We consider the refactoring of currying and its counterpart, uncurrying, using a small proof-of-concept functional programming language with intrinsically typed terms and de Bruijn indices. Furthermore, we proved the retained properties of well-typedness, termination, and value relation of the input program. Finally, we argue that this research could be extended to the full implementation of Haskell, as well as applied to other languages and similar refactorings.

This paper provides a refactoring from do notation to >>= operators and proves that this refactoring maintains observational equivalence. As programs grow ever larger and more complex, there is a growing need to automatically apply refactorings to these programs in a dependable manner. Current refactoring engines often contain errors, even if they are widely used and thoroughly tested. The methods used in the proof of contextual equivalence are described in this paper, with the goal of supporting future research into correct-by-construction refactorings.