Open Radio Access Networks (O-RAN) offer a disaggregated, open alternative to the inflexible and monolithic design of current RAN architectures. The inherent result of the more open O-RAN architecture is an enlarged attack surface, making security a critical focus.

In this paper, the OpenAirInterface and OCUDU test beds, together with the FlexRIC and ORAN-SC near-RT RICs and the OAI-CN5G and Open5GS 5G core networks, are evaluated for vulnerabilities. By exploiting the unencrypted SCTP communications, it was possible to perform a masquerading attack that spoofs O-RAN components, a shutdown attack that injects shutdown messages to stop connections, and a heartbeat attack injecting heartbeat messages with malicious payloads into O-RAN connections.

The attacks found in this research, in combination with the characteristics of the SCTP protocol used in the O-RAN network, allow for the breaking of connections between O-RAN components internally, between network functions in the 5G core, and the near-RT RIC. The breaking of these connections triggers implementation vulnerabilities in the OpenAirInterface and OCUDU test beds, as well as the FlexRIC and ORAN-SC near-RT RICs. These vulnerabilities cause the components to crash or communications between them to be severed permanently.

To secure the O-RAN architecture, strict compliance with the O-RAN specification by implementing IPSec is necessary, together with robust error handling that can safely manage dropped connections and failed connection setup.
... Read more

Modern Internet routing gives end users little control over the paths their packets take and little evidence that packets followed an intended route after transmission. Although routing protocols such as BGP determine reachability at Internet scale, they do not provide packet-level guarantees that traffic traversed a chosen set of routers or avoided untrusted network regions. This gap is increasingly relevant for use cases involving regulatory compliance, data-residency requirements, security-sensitive communication, and post-incident forensic analysis. Existing approaches to path validation and source-controlled routing either require clean-slate deployment, rely on cryptographic primitives that are too expensive for programmable data planes, or introduce per-hop header overhead that grows with path length. This thesis investigates whether controllable packet forwarding can be combined with lightweight cryptographic path verification in programmable data planes. It presents Hermes, a prototype system that allows a sender to route packets through an ordered set of trusted switches and later verify that the packet traversed exactly that path. Hermes uses a compact in-band accumulator carried in each packet. Each switch on the authorised path shares a pool of secret keys and opcode assignments with a central Hermes server. As a packet passes through a switch, the switch applies a keyed operation to the accumulator using only P4-compatible arithmetic and bitwise operations. The receiver emits the resulting accumulator to the control plane, after which the Hermes server independently replays the computation and returns an "Accept" or "Reject" verdict. The design avoids data-plane AES, HMAC, modular exponentiation, and other primitives that are difficult to express in a P4 match-action pipeline. Instead, Hermes uses an optical Diffie--Hellman variant based on XOR and AND operations to provision per-switch key material, and a Galois Linear Feedback Shift Register (LFSR) to drive key-index selection after the initial key pool has been consumed. Replay protection is provided through per-flow sequence numbers, millisecond-resolution timestamps, and server-issued nonces. The prototype is implemented using P4, Python, and C++, and evaluated through a set of experiments measuring key-exchange latency, end-to-end verification latency, throughput and loss behaviour, key-rotation overhead, path-deviation detection, and header overhead. The evaluation shows that Hermes can perform key provisioning within millisecond-scale latency, verify packets with low and stable end-to-end latency across the tested probe rates, and detect the evaluated path-deviation and replay scenarios with perfect accuracy in the experimental setup. The protocol introduces a fixed per-packet header overhead rather than overhead that grows with path length. The security analysis shows that the accumulator construction provides practical path-integrity evidence under the stated threat model, but also identifies important limitations: repeated observations under key reuse remain the dominant source of cryptanalytic risk, the 32-bit accumulator limits the strength of the construction, and the key-exchange channel must be protected in a production deployment. Overall, Hermes demonstrates that programmable data planes can support a lightweight form of controllable and verifiable routing without relying on heavyweight cryptography in the forwarding path. The system should not be interpreted as a replacement for full Internet-scale architectures such as SCION or ICING; rather, it shows that switch-level path verification is feasible as an incremental building block for more accountable network infrastructures. ... Read more

The Network Time Protocol (NTP) is the primary synchronization protocol for billions of devices. Although attacks on NTP servers have been well studied, attacks on NTP clients are less well understood. In this thesis, we scrutinize the 8 NTP clients across 3 Operating systems and test their resilience against attackers and malicious public time servers. We test all the clients against time manipulation, or time-shift attacks and exploit NTP protocol features to induce a denial-of-service for the client by cutting them off from legitimate time servers. We also analyze how these clients perform in normal operations or do they deviate from the NTP-specification-recommended behavior.

Our research analysis unveils 5 bugs and one vulnerabilities, which we disclose to the vendors. While most of the NTP clients follow the associated NTP standards, there exists a wide diversity in the behavior of all the time clients. While many clients follow the specification and best practices, deviations exist which make some of our attacks successful. In detail, macOS time client
is vulnerable to Time Shift attacks, and NTPD-RS is vulnerable to certain Kiss-of-Death packets which causes it to flood the network with queries and OpenNTPd cannot handle the NTP timestamp rollover. We dive into the design and configuration decisions that make the clients vulnerable to different
attacks, we comment of the potential harm, and we propose mitigation. ... Read more

In modern society, critical operations, such as emergency response and public safety, rely on communication systems, in this context also referred to as mission critical systems. These systems must meet strict availability requirements, since any failure can lead to severe consequences, including loss of life. Traditionally, dedicated private communication systems, like local Push-to-Talk systems, were used for such critical operations, but there is a growing shift towards utilizing public 4G and 5G mobile networks to achieve better coverage, higher data speeds and more innovative features at a lower cost.

With the increasing complexity and vast amount of data from the network, automation and artificial intelligence (AI) are now becoming essential tools in these communication systems to efficiently manage data, configure networks in real-time, and effectively handle alarms. The use of AI can improve the end-to-end availability of mission critical systems, ensuring communication during critical situations.

The main goal of this research is to investigate whether and how the use of AI can improve the end-to-end availability of mission critical systems, with a specific focus on the Mission Critical Push-to-Talk (MCPTT) system of KPN, which is using the public 4G and 5G network. Currently, the KPN MCPTT system is being used with a relatively limited number of users. However, the vision for MCPTT extends beyond its current implementation, aiming to scale up this service. With an increasing number of users, using automation and AI is essential for optimizing and managing the complexities of this mission critical communication system.

The implementation of AI in the MCPTT system follows a systematic approach, starting with the independent analysis and monitoring of system specific elements. By focusing on these elements and using data such as Call Detail Records (CDR) and log data, insights into the system's behavior can be obtained. Through collaboration with system experts, AI algorithms can be trained to effectively detect anomalies, thereby enhancing the overall availability of the MCPTT system. Looking ahead, the integration of real-time data becomes crucial for proactive monitoring. Establishing a streamlined data pipeline facilitates the flow of real-time information, offering a comprehensive overview of system performance and enabling swift anomaly detection. It is concluded that the monitoring of individual system elements with the use of AI is a first step towards improving the end-to-end availability.

In order to ensure the correct use of AI throughout the complete cycle, it is crucial to look at explainability, safety, and data quality. These points should be included at each stage of the AI process. By ensuring explainability, system experts can gain insights into the decision-making process of the AI algorithms. By using safety mechanisms, potential risks and vulnerabilities can be mitigated. Maintaining data quality is essential to achieve accurate outcomes. ... Read more

Cloud datacenters underpin our increasingly digital society, serving stakeholders across industry, government, and academia. These stakeholders have come to expect reliable operation and high quality of service, yet demand low cost, high scalability, and corporate (environmental) responsibility. Datacenter operators are confronted frequently with highly complex decisions that involve numerous aspects of risk. The consequence of bad decisions can be financial penalties or even loss of customers on the one hand, or a competitive disadvantage or unsustainable environmental impact on the other hand. Despite risk analysis being an integral part of the design and operation of cloud infrastructure, relatively few comprehensive approaches and tools exist, leaving many datacenter operators ill-equipped to make informed decisions with confidence.

We propose Radice, an instrument for data-driven analysis of IT-related operational risks in sustainable cloud datacenters. Unlike most state-of-the-art approaches used by the industry, Radice automates the process of risk analysis in datacenters and utilizes the large and diverse volume of data reported by the monitoring systems in datacenters, including environmental data. Underpinning this system is the trace-based, discrete-event simulator OpenDC, which enables the exploration of many risk scenarios through its support for diverse workloads, datacenter topologies, and operational phenomena. Radice’s interactive and explorative user interface assists datacenter operators in addressing complex decisions involving risks, providing them with actionable insights, automated visualizations, and suggestions to reduce risk.

We implement Radice and conduct a comprehensive evaluation of the system to demonstrate how it can aid datacenter operators when confronted with fundamental risk trade-offs. Although Radice is designed to work across many kinds of datacenters, in this work, we focus on private-cloud, business-critical workloads, and on public-cloud operations, representing the majority of workloads in Dutch datacenters. Our experiments show many interesting findings, supporting our claim for a need for data-driven risk analysis in datacenters. We highlight the increasing risk faced by datacenter operators due to price surges in the electricity and CO2 bond markets, and demonstrate how Radice can be used to control such risks. We further show that Radice can automatically optimize topology and operational settings in datacenters for risk, revealing configurations that reduce the overall risk by 10%–30%. Following extensive performance engineering, Radice is able to evaluate risk scenarios by a factor 70x–330x faster than others, opening possibilities for interactive risk exploration. We release Radice as free and open-source software for the community to inspect and re-use. ... Read more

Natural disasters can significantly disturb communication networks. There are examples of events causing massive connectivity failures in the past, such as the Great East Japan Earthquake. Network protection mechanisms have been developed to cope with the destructive power of natural disasters and mitigate their impact on connections availability, but in terms of accuracy, they are far from perfect.

Some natural disasters are predictable and can be detected hours or even days in advance. In that case, an adequate protection strategy can be applied. Nevertheless, other types of disasters, such as earthquakes, are classified as unpredictable; thus, protecting the network becomes challenging. Fortunately, early warning systems can detect ground motion and provide a few seconds of warning before the shaking is actually felt. In our work, we utilize early warnings and other disaster data to develop a network protection approach against earthquakes, which operates under rigorous time constraints. Our goal is to minimize the number of disrupted connections in the network by rerouting as many connections as possible out of the disaster zone, such that their availability is maximized. At the same time, the sum of the bandwidth of the connections in the network is also maximized.

We create a realistic disaster model using an early warning system and disaster information. We tackle the uncertainties related to unpredictable disasters by introducing the concept of multiple disaster scenarios. We define the problem of finding paths with maximized availability considering the multiple scenarios. The problem is extended further by adding bandwidth constraints. We propose heuristics to solve the formulated problems and provide an SDN implementation. We validate the effectiveness of our solutions by conducting a series of experiments and creating a custom metric to evaluate our results. The results show that our approach improves the availability of the endangered connections; using the proposed multi-scenario strategy is more beneficial than a single scenario. The results also show that our bandwidth algorithm can optimize the bandwidth utilization of the network. Finally, we compare our solution to an exact solution and find out that our results are very close to optimal. This work provides a mechanism for network operators to ensure the protection of critical network communication in the event of a natural disaster and prevent the potential loss of human lives. ... Read more

Natural disasters can destroy communication network components, potentially leading to severe losses in connectivity. During those devastating events, network connectivity is crucial for rescue teams as well as anyone in need of assistance. Therefore, swift network restoration following a disaster is vital. However, post-disaster network recovery efforts have been proven to be too slow in the past.

Rapidly deployable networks (RDN) are communication networks that can be configured as a wireless mesh network and can be integrated into an existing communication network. As the name suggests, RDNs have a quick setup time and are highly transportable. The technologies behind RDNs for communication networks have received considerable advancements in recent times. Nonetheless, the deployment strategy of such a network remains open.

The existing solutions on rapid post-disaster network recovery are built in an inflexible way. First, each of them is designed around a specific problem. Making slight modifications to the problem greatly increases the complexity of the algorithm and can require major design changes to the system. Second, the proposed solutions are unable to adapt to unexpected circumstances, such as repair times taking longer than anticipated. We propose an online network recovery approach to solve these flexibility issues.

With the optimization objective of maximizing a network's weighted connectivity while minimizing the overall recovery process duration, we design a Deep Reinforcement Learning (DRL) system to produce optimal RDN deployment decisions. Experiments show our Deep Q-network (DQN) algorithm outperform greedy and naive approaches on any disaster scenario. ... Read more

Multi-label learning is an emerging extension of the multi-class classification where an image contains multiple labels. Not only acquiring a clean and fully labeled dataset in multi-label learning is extremely expensive, but also many of the actual labels are corrupted or missing due to the automated or non-expert annotation techniques. Noisy label data decrease the prediction performance drastically. In this paper, we propose a novel Gold Asymmetric Loss Correction with Single-Label Regulators (GALC-SLR) that operates robust against noisy labels. GALC-SLR estimates the noise confusion matrix using single-label samples, then constructs an asymmetric loss correction via estimated confusion matrix to avoid overfitting to the noisy labels. Empirical results show that our method outperforms the state-of-the-art original asymmetric loss multi-label classifier under all corruption levels, showing mean average precision improvement up to 28.67\% on a real-world dataset of MS-COCO, yielding a better generalization of the unseen data and increased prediction performance. ... Read more

Creating content in the digital era has never beenso easy. However, there exists no framework inwhich artists can share, cooperate and transferuniversal content. We refer to the termuniversalas that allowing our architecture to manage contentboth from the physical and digital world. Currentplatforms are either centralized and sector-specificsuch as Netflix and Spotify, or decentralized.On the one hand, the centralized platforms havecontrol over both the data and the identity of theirusers. On the other hand, decentralized platformslack scalability and an identity layer. Our architec-ture expands current decentralized solutions witha more scalable and generic framework. This so-lution allows users, and their verified identities, topossess control over their data; share and exchangecontent, both from the physical and digital world;and get attribution and royalties for their work inany digital coin. ... Read more

Multi-label learning is one of the hot problems in the field of machine learning. The deep neural networks used to solve it could be quite complex and have a huge capacity. This enormous capacity, however, could also be a negative, as they tend to eventually overfit the undesirable features of the data. One such feature presented in the real-world datasets is imperfect labels. A particularly common type of label imperfection is called weak labels. This corruption is characterized not only by the presence of all relevant labels but also by the addition of some irrelevant ones. In this paper, a novel method, Co-ASL, is introduced to deal with the label noise in multi-label datasets. It combines the state-of-the-art approach for multi-label learning, ASL, with the famous robust training strategy, Co-teaching. The performance of the method is then evaluated on noisy versions of MS-COCO to show the lack of overfitting and the performance improvement over the non-robust multi-label ASL. ... Read more

Multi-label learning is becoming more and moreimportant as real-world data often contains multi-ple labels. The dataset used for learning such aclassifier is of great importance. Acquiring a cor-rectly labelled dataset is however a difficult task.Active learning is a method which can, given anoisy dataset, identify important instances for anexpert to label. This greatly reduces the amountof instances needed to train an accurate classi-fier, and thus reduces the cost of cleaning a noisydataset. Therefore, this paper aims to present an ac-tive learning algorithm, focused on wrongly labeleddata, combined with a deep neural network formulti-label image classification. The proposed ac-tive learning solution is divided into two measures;a mislabelling likelihood and an informativenessmeasure together with an option to identify anduse highly probable clean instances in the dataset.Experiments performed on the real world dataset,called Microsoft COCO, with 20, 40 and 60% in-jected label noise show that Multi-AL outperformsthe current state-of-the-art multi-label learning al-gorithm called ASL by 28% while only using 600labelled instances in total and 250 extracted ’clean’instances. Multi-AL additionally outperforms ran-dom sampling by 3% on average for 20 and 40%random label noise when sampling from a wronglylabelled dataset of 23k instances. ... Read more

Digital identity management has been established in a mainly centralized manner. In response to a lack of control in current identity management systems, the concept of Self-Sovereign Identity (SSI) was defined to enable decentralization. Recently, this concept gained traction, and several implementations have been proposed. The decentralized nature of blockchain technology was combined with the concept of SSI. However, no critical review on the privacy protection of this technology in combination with SSI currently exists. This research evaluates current blockchain-based SSI implementations in the lights of privacy protection. It proposes a model for determining the privacy protection that specific solutions can offer based on defined criteria. The technology to be able to satisfy these privacy criteria in blockchain-based SSI is available. However, the evaluation shows that most implementations do not satisfy all privacy criteria, of which some even score poorly on privacy protection. ... Read more

Self-sovereign identity (SSI) provides users of the internet control over their own data by letting them store it on their own device or in a decentralized way, such as on a blockchain. The Super App is an SSI application currently under development by the Delft Blockchain Lab, but it still lacks one of the core features of SSI, which is interoperability. In SSI applications, the user will be in control over their identity when an issuer attests to it. Services can request confirmation about the identity of a user through a verifiable claim, to which the user can reply with this attestation. This research first focuses on building a claim portability framework, which means these verifiable claims and attestations can be communicated between the Super App and other applications. This framework is designed using a public key infrastructure, as that is already present in the Super App. Before sending a claim or attestation, it is signed by the sender and encrypted with the public key of the intended receiver for security purposes. The Super App currently lacks infrastructure to assign issuers of attestations, so a Trusted Issuer registry will have to be stored somewhere in the network. To contest the adoption problem that currently exists in many SSI solutions, the usability has been evaluated as it plays a significant part in adoption. For this, some mock-up user interfaces were created and evaluated by users through a survey and some suggestions were made for improvements. ... Read more

Visible Light Communication (VLC) has gained popularity due to its inherent security as a complementary technology to Radio Frequency (RF) in the last decades to solve the "spectrum crunch'' problem. Meanwhile, The latest IEEE 802.11ah standard, also called WiFi HaLow, offers the range, throughput, and low power consumption that is extremely suitable for most simple IoT appliances for industrial, agricultural, and smart city environments. In general, these IoT products are connected in huge numbers. Hence, provisioning these simple IoT products that usually do not have any user interface like a keyboard or a display in a simple, robust, secure, and scalable method is a significant challenge.

VLC technology has been intriguing both industry and academia for connecting IoT products over the last few years. The signals used in VLC, the visible light, can be captured by eyes and be confined by walls and other blockages which introduce the security against eavesdropping. Besides, the features of low deployment cost, high throughput and high security make VLC a solution to provision IoT products securely. This project exploits on-device existing RGB LEDs to achieve a low-cost and secure device provision system called Integrated Visible Light Communication and Sensing (I-VLCS) System with the functionality of the integrated communication and sensing with RGB LEDs. Visible Light Positioning (VLP), a subset of Visible Light Sensing, is also used in the I-VLCS system to further improve the system security. The performance of the proposed I-VLCS system is evaluated through experiments, demonstrating that that the system can support a maximum 50 cm of communication range with a high communication and positioning accuracy. ... Read more

The accelerating growth of the Internet of Things (IoT) has led to the development of many technologies for homes like a wireless lighting system. Unlike traditional lighting systems, these can be operated remotely using mobile devices and even with the help of a smart assistant, such as Amazon Alexa or Google Assistant. Zigbee is used as the standard of communication between the wireless lighting system whereas other Wi-Fi applications are also extensively used in homes. A common problem for wireless radios such as Wi-Fi and Zigbee is that they have to share the ISM band which could lead to coexistence issues. Zigbee has lower power and longer channel sense times than Wi-Fi which makes it more vulnerable to interference. Bandwidth hungry Wi-Fi applications such as multimedia streaming cause adverse effects on Zigbee’s performance and the system can become temporarily out of reach as the nodes cannot gain access to the channel. This results in inconvenience to the user as the system might not respond in such situations. As these standards are fundamentally different and it is almost impossible to operate on a non-overlapping channel especially with the wide deployment of Wi-Fi networks, a new centralized approach is proposed in this work to coordi- nate between the two heterogeneous networks. The Wi-Fi router which serves as a gateway for both networks in a home is utilized as a coordinator for these networks. By performing various experiments, we examine the lowest data rate for Wi-Fi at which Zigbee can transmit its packet in a reliable manner. We propose a system design that effectively converts a high Wi-Fi interference environment to a transient low interference environment during Zigbee transmission. By combining the packet detection capabilities of the Linux router firewall and a custom queueing setup, we show we can provide reliability to the wireless lighting system with zero to a minimum loss for Wi-Fi transmission. We detect the Zigbee packet well in advance which enables us to adopt a preventive approach rather than a reactive one. Our system design results in a decrease of Zigbee packet loss from 67% to 7% while maintaining an average RTT of 35 ms at every load. We keep the complexity of the system design low by only making software changes to the router and not introducing a new node for synchronization between the two networks. The performance analysis of the system design is done using a test bed consisting of multiple Zigbee and Wi-Fi nodes, with the Wi-Fi router acting as a central controller for both of these networks. ... Read more

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.
... Read more

While multiprocessor platforms have been widely adopted by the embedded systems industry in the past couple of years, there are still fundamental challenges about their timing predictability for applications with real-time timing constraints. The common-off-the-shelf (COTS) multiprocessor platforms typically use complex hardware components, interconnects and multi-level caches which are designed to deliver higher average-case performance. However, these features negatively impact the worst-case performance as they increase the interference of tasks on shared hardware resources. One effective software-based solution to counteract these issues is to use the non-preemptive execution model.

Despite its positive impact on timing predictability, non-preemptive execution causes a potential blocking problem which can decrease the ability to guarantee all timing constraints of the system. It is also known that scheduling non-preemptive periodic tasks on multiprocessor platforms is an NP-hard problem.

In this thesis, we focus on non-preemptive execution of sequential as well as parallel real-time tasks upon multiprocessor platforms and investigate, extend, and improve the state of the art on global, partitioned, and semi-partitioned scheduling approaches for the problem.

We provide the first necessary test for partition-ability, i.e., a test that can determine whether a given task set cannot be partitioned on a given number of cores regardless of the partitioning policy. This test allows us to quantify the pessimism of the existing partitioning heuristics as well as obtain the limits of partitioned scheduling.
We further introduce the first non-work-conserving global scheduling policy and show that despite the fact that it improves over the existing global scheduling policies, it is not as effective as the partitioned scheduling strategies.

We extend a sustainable scheduling algorithm designed for uni-processor platforms to multiprocessor ones to improve the performance of partitioning heuristics. A sustainable scheduling algorithm does not have timing anomalies and hence it is easier to analyze and can have better scheduling results.

Furthermore, we introduce the first semi-partitioned non-preemptive scheduling solution for multiprocessor platforms. Our solution is able to schedule some of the task sets for which it is impossible to find a partitioning solution. Finally, we compare the overheads and memory consumption of various scheduling approaches (including ours) on a bare-metal multiprocessor hardware platform, i.e., a 4 processor Raspberry Pi board. We show that our sustainable scheduler has a very low overhead while it out-performs other solutions in terms of schedulability. ... Read more

Large cities in the Netherlands, like Rotterdam, have hundreds of playgrounds, but local governments have little information on how children and adults use them. Usage data can help create playgrounds that better fit with the residents' needs by identifying what elements of a playground are the most popular, and which do not get used. The AMS Institute and municipality of Rotterdam have asked us to develop a system to collect usage information of playgrounds without recording personally identifiable information. The system requirements are to record the locations of individual users and to estimate if a user is a child or an adult. It must do this without requiring an external power source and without a broadband internet connection. After considering different types of sensors (including computer vision, radio, sound, and mechanical acoustic signals), We decided to use a mmWave radar sensor due to its ability to provide accurate localization, easy installation, and low energy usage without recording identifiable information. We use a commercially available mmWave radar sensor that we configure to localize people in a 30m by 20m area when placed at the perimeter. We use the radar point cloud output from the device for classification by calculating statistics that we use as features for our classifier. We evaluate classifiers based on SVM, Random Forrest, fully connected and recurrent neural nets. We also analyze different methods for combining radar point clouds captured over time from the same person. We collected 100.000 radar point clouds of adults, children, and bicyclists at real playgrounds, split into a training and validation data set. We show that our SVM classifier achieves an accuracy of 79% on single radar frames, and 93% on the combined results of 10 second long sequences of our validation dataset. The classifier requires <1KB of memory and little processing power. Meaning it can execute on an embedded platform powered by a solar panel. ... Read more

Open-domain question answering (QA) is an important step in Artificial Intelligence and its ultimate goal is to build a QA system that can answer any question posed by humans. The majority of the open-domain QA system is the retrieval-based open-domain QA system, which enables the retrieval component to retrieve relevant documents from a large-scale knowledge source to a question and the answer extraction component to extract the answer to this question based on retrieved documents. As the techniques of Deep Learning progressing significantly, many researchers tried to apply the neural reading comprehension (RC) model to serve the answer extraction component of the open-domain QA system. However, the performance of the neural RC model in open-domain QA is considerably worse than the performance of it in RC-style QA. Therefore, many works have focused on the neural RC model for addressing the performance gap, whereas the retrieval component of the open-domain QA system lacks equivalent attention. Some researchers have built the neural network based information retrieval (IR) models, but currently, it is still difficult for these neural IR models to directly retrieve documents from a large-scale knowledge source in open-domain QA. Hence, many works attempted to use neural IR models for re-ranking documents retrieved by the traditional but efficient IR models (e.g., TF-IDF, BM25) in open-domain QA. However, these works did not analyze the impact of different questions of QA datasets on traditional IR models. Thus, this research gap is the focus in this thesis. We conduct error analyses of questions of different QA datasets to figure out the error types of questions that have a negative impact on the traditional IR models. From the error analysis, we learn that different QA datasets have different impacts on the traditional IR models and are differently hard to be dealt with by the traditional IR models. Therefore, we propose hypotheses that might mitigate the negative impact of the error types of questions that are relatively harder to be handled by the traditional IR models. Furthermore, we perform experiments based on the methodologies that implement our hypotheses for figuring out the validity of these hypotheses. In conclusion, we believe that our work is a step forward to obtaining more insights into the retrieval component of the open-domain QA system and will contribute to the development of the retrieval component for a better open-domain QA system. Moreover, our work can give our users guidance on how to issue a more suitable question that can be processed by the open-domain QA system for giving a more accurate and better answer. ... Read more

It is predicted that in about 100 years most of the earth's fossil fuels will have been depleted. Currently, fossil fuels still make up 80% of the Dutch energy production. Thus, to handle this depletion, the research on renewable energy production and its usage is being stimulated by governments. With the use of fossil fuel energy production, it is easy to increase production to meet the unexpected peaks in energy demand by burning more fuels. However, with renewable sources this is not possible. Thus, the way that the produced energy is being used needs to be altered. Furthermore, the amount of renewable energy that is privately generated has increased over the last couple of years. The energy that has been generated for private use and is not needed at that time, can either be sent back into the grid for other users or charged to a battery for later personal use. The electricity network that will regulate the buying and selling of energy is called a smart grid. When using a battery to store privately generated energy, the decisions that are made for the (dis)charging of the battery are of great influence on the total energy cost at the end of the month. When implementing a battery in a household or company that privately generates energy, these decisions need to be made within a fixed time limit of 15 minutes. In this thesis, four mathematical optimisation methods are compared to each other on result and run time. These methods are dynamic programming, local search, tabu search, and simulated annealing. Dynamic programming gives the solution with the lowest possible cost, but does not always have the lowest average run time. The total cost of the solution resulting from local search does not come close enough to the lowest possible cost generated by dynamic programming to be a viable alternative to dynamic programming. Tabu search is an extension of local search, it could result in a solution with a total cost close enough to the lowest possible cost if it runs more iterations than local search. However, due to this the average run time will exceed the run time of dynamic programming. Therefore, it is also not a viable alternative for dynamic programming. Simulated annealing has a shorter run time than dynamic programming when using a forecast time of 1 day or less. The total cost of the solutions come very close to those of dynamic programming. Therefore, while the run time of dynamic programming still fits within the available time limit, it is advisable to use this method to determine the charging decisions for a private battery in a smart grid. However, the simulations that were run in for this thesis do not encompass the entire real-life case. If after the expansion of the problem to be implemented in real-life the run time of dynamic programming were to exceed the available time period, then simulated annealing would be a good alternative for implementation. ... Read more