Label-independent concept drift detectors represent an emerging topic in machine learning research, especially in models deployed in a production environment where obtaining labels can become increasingly difficult and costly. Concept drift refers to unforeseeable changes in the distribution of data streams, which directly impact the performance of a model trained on historical data. This paper initially focuses on two mixed label-independent drift detectors, SQSI and UDetect, which are implemented and evaluated on a specific setup using synthetic and real-world data sets. Next, multiple label-dependent drift detectors are evaluated on real-world data sets, and the results are compared to those of the label-independent detectors. This paper presents a framework for comparing multiple concept drift detectors on different data sets and configurations, checking whether they can be reliably used in a production environment.

At the moment we are witnessing the maturation of distributed streaming dataflow systems whose use-cases have departed from the mere analysis of streaming windows and complex-event processing, as they now extend to cloud applications, workflows and even e-commerce. The state of streaming operators has been so far hidden from external applications. In this thesis it is argued that exposing this internal state to outside applications by making it queryable, opens the road for novel use-cases. To this end, we introduce S-Query: a system and reference architecture where the state of stream operators can be queried - either live or through snapshots, achieving different isolation levels. It is shown how this can be implemented in an existing open-source stream processor and how queryable state can affect the performance of such a system. Our experimental evaluation suggests that snapshot configuration adds only up to 5ms in latency in the 99.99th percentile and no increase in 0-90th percentiles.

Low latency Convolutional Neural Network (CNN) inference research is gaining more and more momentum for tasks such as speech and image classications. This is because CNNs have the ability to surpass human accuracy in classication of images. For improving the measurement setup of gravitational waves, low latency CNNs inference are researched. The CNN needs to process data from images to enable certain automatic controls for the control system. The data of these images need to be processed within 0.1 ms from the moment of taking the image to the control system obtaining the result of a deep neural network. Hardware acceleration is needed to reduce the execution latency of the network and reach the 0.1 ms requirement. Field-Programmable Gate Arrays (FPGAs) in particular have the ability to provide the needed acceleration. This is because FPGAs have the ability to create highly customised layers of the network and obtain the lowest possible latency. To reduce the design eort and complexity of the machine learning design, Xilinx introduced the FINN (Fast, Scalable Quantized Neural Network Inference on FPGAs) framework. FINN is an end-to-end deep learning framework that generates data ow-style architectures customised for each network. To establish if FINN can create the required ultra low latency CNN, some of FINN pretrained networks are used. The rst neural network investigated is the Tiny Fully Connected (TFC) network. The TFC network is a multilayer perceptron (MLP) for MNIST classication with three fully connected layers. The other network investigated is the convolutional neural network named CNV. CNV is a derivative of the VGG16 topology. The VGG16 topology is used for deep learning image classication problems with multiple convolutional layers. By using the analysis tools included with FINN, it can be determined if FINN is able to create the required ultra low latency CNN. The TFC network can be parallelised to a total of 5 expected cycles, with 1 expected cycle per layer. One cycle for the input quantization to standalone thresholding, one for the output layer and nally three for the fully connected layers. For the CNV network on the other hand, the initial convolution layer is unable to go below 8196 expected cycles, because of certain bottlenecks with FINN. These bottlenecks occur because of how FINN implements certain layers, moreover because certain layers can simply no longer be parallelised to lower the latency of that layer. To see if the CNV could achieve the latency requirements, a software emulation of the execution of the network has been done. This emulation showcased that by continuously increasing the parallelisation parameters, together with increasing clock frequencies, it is possible to create an ultra low latency pipeline of a CNN. This conguration has 45866 expected total cycles for the network, its expected cycles for the slowest layer is 8196 and needs a minimum frequency of 200 MHz. With those conguration it is possible to create a pipeline that has a latency of lower than 0.1 ms.

With the increase of available storage bandwidth, CPUs can not keep up with the compute throughput needed to process this amount of incoming data. GPUs and FPGAs are generally better suited for such tasks. To assist FPGAs in their functions, some boards are equipped with one or more high bandwidth memory (HBM) stacks, with a bandwidth of 230 GB/s each. This thesis presents a hardware design for the Alveo U280 FPGA board with HBM. Each HBM stack provides multiple interfaces to the full range of memory within HBM. Utilizing these multiple interfaces, a hardware decompressor for the Snappy compression algorithm is placed in parallel to achieve a higher end-to-end throughput. Additionally a design is created to perform benchmarks on HBM where varying sizes of data are transported between HBM and logic within the FPGA. The hardware decompressor and component that interfaces to memory within HBM were found to be incompatible and required additional logic to become able to transport data between them. To ease the parallelization of the decompressor a custom Snappy framing format is implemented. Using this format a softcore processor on the FPGA is able to buffer the locations of compressed data within HBM and divide these over available decompressors. The design is successfully synthesized into a kernel that can be loaded by the FPGA. From the moment compressed data is sitting in HBM until it is decompressed, a single decompressor reaches a maximum end-to-end throughput of 4.0 GB/s. When eight or more decompressors are activated, they reach a throughput between 20.0 to 26.2 GB/s. The hardware decompressor designs uses less than 10% of the resources of the U280 with little power usage. Compared to a software implementation, using multithreading, the hardware solution is 1.5-2.5x faster on a set of files that is used for benchmarks on decompression speed with varying compression ratios.

Big data applications are becoming more commonplace due to an abundance of digital data and increasingly powerful hardware. One of these classes of hardware devices are FPGAs, which are being used today in various ways such as data centers and embedded systems. High performance, power efficiency, and reprogrammability are the primary reasons behind their wide use. Another trend over the previous years has been to use distributed data processing frameworks such as Apache Spark to improve the performance of big data applications. Traditionally, such frameworks are deployed on commodity hardware to save costs. This approach is fairly popular, with organizations often having on-premise compute clusters or using a cloud provider to access a managed cluster. This project attempts to combine the above-mentioned worlds - FPGAs and dis- tributed data processing. We have designed an architecture that allows us to use FP- GAs as end-devices in a compute cluster to perform the actual computation instead of CPUs. This architecture is designed by composing together several open source technologies and allows us to interact with an FPGA cluster using Python. Using a high-level programming language such as Python makes this system easy to use for software developers and data scientists, and also abstracts away the internal commu- nication within the cluster. We have built prototypes based on this architecture for 3 hardware platforms (FPGA families) and 3 specific applications to demonstrate general applicability. We have observed noticeable performance gains in these applications by scaling up the FPGA cluster.

In the traditional software development life cycle, development and operation are divided into different departments. The conflict between departments and, besides, the lack of automation usually leads to low software development efficiency and slow software delivery. Thus, the concept of DevOps is introduced, which combines different departments and automates the process to make software delivery faster and easier. The DevOps toolchain is one important component for adopting DevOps. On the other hand, the adaptation of cloud technology, especially serverless computing makes it tempting for us to investigate what benefits serverless computing brings to the DevOps toolchain. In the first research question, we examine the benefits that AWS serverless platforms bring to DevOps toolchain. To answer this research question, we develop a DevOps toolchain hosted in Amazon Web Services (AWS) and leverage the serverless computing service. In addition, we examine what does each serverless computing service brings to the DevOps toolchain, examine how does the performance of the DevOps toolchain changes with or without using serverless computing service. Our research shows that serverless computing services such as AWS Fargate could reduce the cost, operation effort, and improves performance by enabling parallel execution. Our experiments show that in contrast to a toolchain hosted in a traditional cloud server vs the toolchain that was developed by us using serverless computing service could reduce the total runtime of parallel execution up to 65%. In the second research question, we focus on the integrated toolchain build with AWS DevOps tools from AWS serverless platform. We build a demo integrated DevOps toolchain with AWS DevOps tools and compare the integrated toolchain with the non-integrated toolchain that we built. We find that the integrated toolchain significantly reduces the development time by providing an out-of-box solution for the software team. In addition, the better integration with underlying cloud infrastructure provides more functionality such as global monitoring and blue/green deployment. However, we also find that from the experiment that the performance of the integrated toolchain is lower due to the limitation of resources which also come with a high cost.

With the growth in the complexity of real-time embedded systems, there is an increasing need for tools and techniques to understand and compare the observed runtime behavior of a system with the expected one. Since many realtime applications require periodic interactions with the environment, one of the fundamental problems in guaranteeing/monitoring their temporal correctness is to be able to infer the periodicity of certain events in the system. The practicability of a period inference tool, however, depends on both its accuracy and robustness (resilience) against noise in the output trace of the system, e.g., when the system trace is impacted by events that have a non-deterministic nature such as the presence of aperiodic tasks, release jitters and runtime execution-time variations of the tasks.
This work (i) presents a period inference framework that uses regression-based machine-learning (RBML) methods, (ii) thoroughly investigates the accuracy and robustness of different families of RBML methods in the presence of uncertainties in the system parameters, and (iii) proposes further accuracy improvements by deriving candidate pruning rules based on the inherent properties of the underlying scheduling policies. We show, on both synthetically generated traces and traces from actual systems, that our solutions can reduce the error of period estimation by two to three orders of magnitudes w.r.t. state of the art. Also, our methods showed to be robust against most sources of disturbance.

Hyperparameter optimization(HPO) forms a critical aspect for machine learning applications to attain superior performance. BOHB (Bayesian Optimization and HyperBand) is a state of the art HPO algorithm that approaches HPO in a multi-armed bandit strategy, augmented with Bayesian optimization to drive configuration sampling. However, BOHB requires predefined distribution of fidelities for each tuning task. The challenge in this is that it is impossible to define fidelities a priori, since each machine learning model is uniquely complex and requires different amount of compute resources for convergence. Furthermore, in our empirical analysis, we found that each HPO task rendered different performance trajectories on different fidelity (budget) types. Thus, the challenge of defining fidelities also extends to choosing an optimal budget type. To alleviate these challenges, we present MultiTune: a budget allocation scheme that builds on top of BOHB to dynamically define fidelities for optimization. MultiTune incorporates an algorithm to dynamically choose a preferred budget type for an HPO task, coupled with 2D gradient based budget constraint explorations to enable granular definition of fidelities. Through our empirical analysis, we show that MultiTune can consistently converge to a well performing configuration without significant computation overhead.

Systematic literature reviews in software engineering as well as other disciplines, serve as the foundation for sound scientific research. The aim for these literature reviews is to aggregate all existing knowledge on a research problem and produce informed guidelines for practitioners. This enables practitioners to apply appropriate software engineering solutions in a specific contexts. However, one major problem exists regarding systematic literature reviews, the overall execution duration may take up as much as 24 months. The first objective of this study is to provide a solid base for the AI for FinTech Research collaboration by performing a systematic literature review. This literature review is used to identify different machine learning techniques in the context of the FinTech domain. However, during this study, we found that a significant amount of time was spent on repetitive work which potentially could have been automated. Therefore, the second objective of this work is to reduce the overall workload for performing systematic literature reviews. First, a literature review is performed regarding automation solutions for different steps of systematic literature reviews. The identified solutions were used to create a tool to automate steps in both the retrieval and screening phase of systematic literature reviews. First, this work presented the state of the art regarding machine learning applications in the FinTech domain. Afterwards, a complete overview of possible automation solutions for every step of performing literature reviews was detailed. Using this overview, a tool was created which showed that the overall workload of the retrieval and screening phase of systematic literature reviews can be significantly reduced.

Self-Sovereign Identity (SSI) is a new paradigm in digital identity systems that puts the end-user in control: no other actor manages, permits or revokes their digital existence. TrustChain is an academic peer-to-peer networking stack supporting SSI. It delivers passport-grade assurance by integrating with Dutch government. However, end-user control requires a programmed user agent with a human interface and protocols that enable meaningful communication with issuers and verifiers of identity data. This agent must be inter-operable with a large variety of parties and credentials. TrustChain lacks such an interface and protocols.   This thesis makes three main contributions. First, a theoretical framework is proposed for aligning notions of self-sovereignty across contexts, borders and cultures. It provides more detailed, focused and structured discourse than other work and helps consolidate design efforts. Second, a design project is done in collaboration with the Kamer van Koophandel (KVK). It focuses on `authorisation by legal entities', a class of identity problems that have no satisfactory solution yet. Third, a generic common `semantic layer' is prototyped, consisting of a smartphone based user agent and communication protocols. Its wallet-centric approach allows end-users to retrieve their data without leaving the app. The practical value of this prototype is evaluated at a construction site.  The case study shows that the Kamer van Koophandel, like other government institutions, can be a valuable data provider. However, their current legal framework and business model may restrict them. Absence of such vital institutions invites commercial parties to close the gap, threatening privacy and independence of end-users.   Finally, this work has three implications for TrustChain. First, attestation metadata must be considered confidential. Second, single-sided public revocation is required to ensure credential actuality without re-issuing. And third, non-interactive verification enables the construction of chains of untrusted issuers. This is a valuable feature as it enables individuals, not just organisations, to issue claims to others.  

With the increasing amount of container freight transport and the increasing size of container vessels, for the Port of Antwerp, the second largest container port in Europe, a critical task is port planning. A simulation model provides the means to gain proper insight in the effect of future expansions. Macomi, a company specialized in simulation and optimisation, has been working on a simulation model to aid the Port of Antwerp in their port planning. One important issue in this simulation model is how the berth allocation of vessels is handled. Berth allocation is the problem of assigning vessels a time and location at the quay wall where the vessel can be loaded and unloaded. In this thesis, the aim is to develop decision models for both the preliminary berth planning and the real-time recovery of this plan during simulation. For the first part, a cyclic baseline berth allocation plan is created which takes into account the tidal dependencies vessels have when entering the port of Antwerp. This preliminary berth plan is used as a basis for the simulation model as the arrival times are based on this plan. However, during the simulation disruptions might occur; vessels can arrive earlier or later or take longer to load and unload. To deal with these disruptions a real-time disruption management decision model is proposed which aims to solve all disruptions while staying as close to the theoretical berth plan as possible. Using the proposed models, several experiments have been conducted regarding the influence of uncertainty, occupation and robustness on the quality of the solutions that the decision models found. Regarding occupancy rates, results show that a tipping point exists where the recovery model has more difficulty to find a good solution. Results also show that when the expected occupation of a terminal is higher, adding robustness has more effect and is therefore more important. The decision models presented in this thesis have been implemented in the Macomi port simulation model and have been demonstrated to the Port of Antwerp. Both parties have expressed their satisfaction with the models.

Recently, there is a growing need to address the investment gap in renewable energy by attracting non-institutional investors that have investment preferences better suited for the energy transition. Crowdfinancing is a promising alternative, as it allows for increased innovation, more sustainable development and reduced community risks and is driven by a crowd of small investors that are less profit-driven than professional investors. However, current crowdfinancing models are limited by lack of trust, adverse selection, exclusive corporate governance and lack of liquidity. While extant research has identified these limitations and the potential of blockchain technology to solve some of them, there is a need to design the blockchain-based crowdfinancing mechanisms required and provide a business case for their implementation. The research objective of this practice-oriented design research is to provide a practical plan for the implementation of blockchain-based crowdfinancing for renewable energy projects by means of a design and business case for engineering and consultancy firms as the facilitators of the process in the Netherlands, with the purpose to give a boost to the energy transition.
A foundation for the research was provided by making an analysis of the evolution and current state of the Dutch energy market organization based on three cornerstones: market roles, production and financing. This analysis showed a high degree of decentralization in terms of market roles and production from renewables that allowed the emergence of community driven energy cooperatives, while a dependence on centralized financial intermediaries is still present; risky and small projects do not have access to bank financing and existing crowdfinancing platforms charge high brokerage fees and provide no secondary markets. A pilot study based on the same cornerstones and consisting of interviews with relevant market parties was performed in order to supplement the analysis of the market for the formulation of design requirements for the crowdfinancing mechanisms. Qualitative desk research was done on existing smart contract applications that could be utilized to overcome the limitations of crowdfinancing and served as the building blocks for the conceptual designs. Qualitative desk research was also performed in order to quantify the costs and address the (legal) implications of implementing the mechanisms proposed.
The result was a practical plan consisting of blockchain-based crowdfinancing mechanisms that reduce brokerage fees and facilitate the organic establishment of secondary markets by making use of existing public blockchain infrastructure, and a positive business case for consultancy and 10 engineering firms to facilitate the process. Smart contracts govern the crowdfinancing process autonomously for low and high risk projects, where with the latter crowdinvestors are given financial control by locking project funds in trustless escrow contracts, thereby allowing them to cut losses in failing projects and thus addressing the issues of lack of trust and exclusive corporate governance, while mitigating the effects of adverse selection. This practical plan provides a tool to further empower the crowd as the fuel for the energy transition, thereby increasing investments in renewables. However, social exclusion might result at first from the intrinsic complexity of blockchain technology and the degree to which a boost can be given to the transition remains unclear due to a lack of consensus on the effectiveness of crowdfinancing to increase support for the bigger projects. There is further research needed into a risk management-based approach for the design of escrow contracts that release funds according to a project planning, while the delegation of control rights to more capable third parties also requires more research. Additionally, the application of decentralized governance beyond financial control requires further examination as well.

For decades the idea of re-usable software has been seen as the holy grail of software development. But up until recently, there was more discussion about software re-use than actual software re-use. Even though most software uses the same blocks of code over and over again, almost all software is built from the ground up. Today, this situation is completely different. Nowadays, almost every application re-uses software in the form of software dependencies. However, this re-use pattern is starting to become unchecked. The shift to re-usable software has happened so quickly, the risks associated with choosing the right dependencies are often overlooked. In the current research, there exists a gap in the balance between re-usability and usability. This work tries to fill that gap. People have tried solving the problem of software re-use, but it has proven to be a hard problem. There needs to be a trade-off between re-usability and usability. This thesis focuses its work on developing a framework that continues the progression in the develop- ment of re-usable code. This is achieved by enabling trustworthy code and providing runtime support. A key property is permission-less code execution at near-zero cost. The concept, called FBase, sets out to achieve the balance between re-usability and usability by limiting the granularity of re-usable modules to a distinct set of four component types. To further enhance the us- ability of FBase, an ecosystem was proposed to mask the negative effects that are associated with re-usability. Previous attempts at solving the re-usability problem have mostly focused on a technical level in contrast to FBase which also incorporates social and policy aspects. By integrating sub-systems into an ecosystem it improves the usability of the framework. A proof-of-concept implementation is created to support the eval- uation of the concept. To be considered successful, FBase needs to perform properly on a technical and functional level. The evaluation showed that FBase can be used in a non-trivial use-case, by creating a fully functioning example that demonstrates the composition and construction of an application with interchangeable trust models. It was demonstrated that real-world practical problems can be solved using this framework. The main advantage that the framework provided in the use-case was the introduction of modularity. On one hand, it has the benefit of flexibility and variety in use. On the other hand, modularization improves the manageability of maintenance for complex software like Tribler. One of the disadvantages that followed from the use-case was the increased time that modules on FBase took to develop the application compared to implementing it in a monolithic architecture. A second more general disadvantage is that the module interconnect limits the complexity of the interaction between mod- ules. This disadvantage did not limit the development of this use-case but changes the way applications need to be developed. To test the robustness and flexibility of the framework, an experiment was performed to try to create a proof-of-concept prototype of an Android application that could run the same stack of code to extend the ecosystem to mobile platforms. FBase delivers a concept that complies with the requirements that were set out and offers a balanced ap- proach of re-usability and usability. In essence, FBase is a platform for building applications consisting of re-usable code. This makes it similar to Python with the Pip dependency manager or NodeJS and its depen- dency manager NPM. However, these existing platforms make use of centralized approaches and do not solve the trust aspect of using external dependencies in contrast to FBase.

Ever since its invention in the 1930s, photoplethysmography (PPG) is a wide-spread technique used for health-monitoring. Via illumination of the human skin with a light source and capturing the light, an estimate of important physiological properties such as the heart rate can be made. This is commonly done with dedicated medical equipment, but studies from the last decade have shown that the smartphone could also be an adequate sensor, using the flash light and camera as transmitter and receiver respectively. Small, mobile and off-the-shelf, the smartphone provides many advantages over the conventional sensor. However, the quality of the PPG signals measured is a point of concern, as these could lead to incorrect estimates and several challenges need to be addressed. Firstly, user movement disturbs the contact area between the skin and sensor, introducing distortions in the PPG signal. Especially subtle motion artifacts like finger contact pressure impact the PPG signal quality. Secondly, the smartphone camera has a wide range of settings that can be used, but a proper analysis was lacking in literature. Thirdly, recent research has shown that PPG signals from an individual are unique. There is no common morphology. This means that algorithms developed for PPG need to account for unknown characteristics of PPG signals. Fourthly, analysis on PPG signals from the smartphone is mainly done offline and as such, a real-time implementation is desired. This thesis introduces a smartphone application that tackles these challenges and provides PPG signals of high quality. A real-time multi-stage PPG qualification pipeline combined with a pulse segmentation algorithm is proposed. Furthermore, analysis of the camera settings and finger detection algorithm resulted in PPG signals with 12.63% higher quality than a dedicated PPG sensor.

Online social networks have revolutionized the way people interact with each other nowadays. Users often share their experiences in various health - related topics like disease symptoms, drug treatments and other medical related issues in order to discuss with other patients dealing with similar conditions. During the production of a new drug, important drug properties like possible Adverse Drug Reactions (ADRs) are monitored through a phase of clinical trials. However, due to various factors that can not be easily measured in those trials, patients can potentially experience adverse events that were not related to their treatment before, or were related to it in a much smaller frequency. Therefore, the automatic detection of Adverse Events in online networks is gaining an increasing popularity among researchers in the biomedical community, as it can offer a valuable complementary source of information, next to the traditional approaches of reporting those events to the corresponding Food and Drug Association. From an NLP perspective, this task poses a significant challenge as there is a large gap between the informal language used in social media and the formal medical language used to officially describe a medical concept. Moreover, there is an absence of large annotated datasets, as the manual labeling of an adverse effect mentions is a time-consuming and often ambiguous procedure which also requires some sort of medical expertise. In this work we propose a novel machine learning approach to normalize Adverse Drug Effect mentions in user-generated text to a standard vocabulary from a medical Ontology. The evaluation results of the proposed model demonstrates a competitive performance among the current state of the art techniques, posing the potential feasibility of our model in the medical concept normalization domain.

vailability of FPGAs is increasing due to cloud service offerings. In the wake of a new in-memory storage format specification, Apache Arrow, FPGAs are increasingly interesting for data processing acceleration in the big data domain. The Fletcher framework can be used to easily develop FPGA accelerated applications that access data stored in Apache Arrow format, while providing throughput near the system’s limit. The current implementation of Fletcher has limited support for memory management of FPGA-local memory, with one of the biggest limitations being that memory can only be used once.
This thesis explores several memory management techniques which could be suitable for use on FPGAs in a big data context. Paged memory is implemented on FPGA within the Fletcher framework in order to facilitate this memory management. The implemented system takes less than 5 % of a data centre FPGA card’s resources (Xilinx UltraScale+ VU9P). Experiments show that the paged memory provides throughput of over 99.7 % of the system’s throughput for linear memory accesses. Random memory access throughput for paged memory drops to between 30 % and 90 % of the system’s original throughput, depending on request size. The performance drop can be lightened or even prevented by employing suitable address-translation caches.