132571 research outputs found
Sort by
Connecting the dots: Exploring backdoor attacks on graph neural networks
No full textDeep Neural Networks (DNNs) have found extensive applications across diverse fields, such as image classification, speech recognition, and natural language processing. However, their susceptibility to various adversarial attacks, notably the backdoor attack, has repeatedly been demonstrated in recent years. The backdoor attack aims to misclassify inputs with specific trigger pattern(s) into the pre-determined label(s) by training the model on the poisoned dataset. Backdoor attacks on DNNs can lead to severe real-world consequences, e.g., a deep leaning-based classifier in a self-driving car can be backdoored to misclassify a stop sign as a speed limit sign. With an increasing of real-world data being represented as graphs, Graph Neural Networks (GNNs), a subset of the DNNs, have demonstrated remarkable performance in processing graph data. Despite their efficiency, GNNs, similar to other DNNs, are also vulnerable to backdoor attacks, which can lead to severe results, especially when GNNs are applied in security-related scenarios. Although backdoor attacks have been extensively studied in the image domain, we still need dedicated efforts for the graph domain due to the difference between graph data and other data, e.g., images. This thesis embarks on an exploration of backdoor attacks on GNNs. Chapter 2 focuses on designing and investigating backdoor attacks on centralized GNNs. Specifically, we explore the influence of trigger injecting position on the backdoor attack performance on GNNs. To explore this impact, we propose approaches based on explanation techniques on GNNs, which contributes to exploring the interaction between the explainability and robustness of GNNs. Furthermore, we design a clean-label backdoor attack on GNNs to make the poisoned inputs more challenging to be detected. Considering the growing privacy concern, we focus on backdoor attacks on federated GNNs in Chapter 3. We propose a label-only membership inference attack on GNNs in the scenario that the attacker can only get label output from the GNN models. Moreover, we investigate centralized and distributed backdoor attacks on federated GNNs. Besides designing efficient backdoor attacks on GNNs, we also explore the possibility of leveraging backdoor attacks for defensive purposes for GNNs. Chapter 4 introduces a watermarking framework for GNNs based on backdoor attacks. Our research outcomes will deepen the understanding of backdoor attacks on GNNs and push the GNN model designers to develop more secure models.Cyber Securit
Radiation resistant metal–organic frameworks for the production of high specific activity 51Cr by the Szilard-Chalmers effect
No full textChromium-51 (51Cr) is an attractive radionuclide for diagnosis, which is usually applied for red cells and platelet radiolabeling. However, commercially available 51Cr produced in nuclear reactors via neutron activation requires long irradiation times and complex separation methods. In this work, five metal–organic frameworks (MIL-100 (Cr), MIL-100 (Fe), MIL-100 (Al), MIL-101 (Cr) and aluminium fumarate MOF (FuAl)) were synthesized and the effect of gamma ray irradiation with a high dose rate and a maximum dose of 6 MGy was investigated. The two chromium-based MOFs, MIL-100 (Cr) and MIL-101 (Cr), were selected as radiation targets to produce high specific activity 51Cr by the Szilard-Chalmers effect. A solid–liquid extraction was applied to extract the produced 51Cr under different conditions, including different extractants, extraction times and pH. The most promising results were achieved when using irradiated MIL-101 (Cr) and EDTA as extracting agent, reaching an enrichment factor of 1132 ± 50.RST/Applied Radiation & Isotope
Is this network proper forest-based?
No full textIn evolutionary biology, networks are becoming increasingly used to represent evolutionary histories for species that have undergone non-treelike or reticulate evolution. Such networks are essentially directed acyclic graphs with a leaf set that corresponds to a collection of species, and in which non-leaf vertices with indegree 1 correspond to speciation events and vertices with indegree greater than 1 correspond to reticulate events such as gene transfer. Recently forest-based networks have been introduced, which are essentially (multi-rooted) networks that can be formed by adding some arcs to a collection of phylogenetic trees (or phylogenetic forest), where each arc is added in such a way that its ends always lie in two different trees in the forest. In this paper, we consider the complexity of deciding whether a given network is proper forest-based, that is, whether it can be formed by adding arcs to some underlying phylogenetic forest which contains the same number of trees as there are roots in the network. More specifically, we show that it is NP-complete to decide whether a tree-child network with m roots is proper forest-based, for each m≥2. Moreover, for binary networks the problem remains NP-complete when m≥3 but becomes polynomial-time solvable for m=2. We also give a fixed parameter tractable (FPT) algorithm, with parameters the maximum outdegree of a vertex, the number of roots, and the number of indegree 2 vertices, for deciding if a semi-binary network is proper forest-based. A key element in proving our results is a new characterization for when a network with m roots is proper forest-based in terms of certain m-colorings.Discrete Mathematics and Optimizatio
Optimally reconfiguring list and correspondence colourings
No full textThe reconfiguration graph Ck(G) for the k-colourings of a graph G has a vertex for each proper k-colouring of G, and two vertices of Ck(G) are adjacent precisely when those k-colourings differ on a single vertex of G. Much work has focused on bounding the maximum value of diamCk(G) over all n-vertex graphs G. We consider the analogous problems for list colourings and for correspondence colourings. We conjecture that if L is a list-assignment for a graph G with |L(v)|≥d(v)+2 for all v∈V(G), then diamCL(G)≤n(G)+μ(G). We also conjecture that if (L,H) is a correspondence cover for a graph G with |L(v)|≥d(v)+2 for all v∈V(G), then diamC(L,H)(G)≤n(G)+τ(G). (Here μ(G) and τ(G) denote the matching number and vertex cover number of G.) For every graph G, we give constructions showing that both conjectures are best possible, which also hints towards an exact form of Cereceda's Conjecture for regular graphs. Our first main result proves the upper bounds (for the list and correspondence versions, respectively) diamCL(G)≤n(G)+2μ(G) and diamC(L,H)(G)≤n(G)+2τ(G). Our second main result proves that both conjectured bounds hold, whenever all v satisfy |L(v)|≥2d(v)+1. We conclude by proving one or both conjectures for various classes of graphs such as complete bipartite graphs, subcubic graphs, cactuses, and graphs with bounded maximum average degree. The full paper can also be found at arxiv.org/abs/2204.07928.Discrete Mathematics and Optimizatio
Less stick more carrot? Increasing the uptake of deposit contract financial incentives for physical activity: A randomized controlled trial
No full textBackground: Financial incentives are a promising tool to help people increase their physical activity, but they are expensive to provide. Deposit contracts are a type of financial incentive in which participants pledge their own money. However, low uptake is a crucial obstacle to the large-scale implementation of deposit contracts. Therefore, we investigated whether (1) matching the deposit 1:1 (doubling what is deposited) and (2) allowing for customizable deposit amounts increased the uptake and short term effectiveness of a deposit contract for physical activity. Methods: In this randomized controlled trial, 137 healthy students (age M = 21.6 years) downloaded a smartphone app that provided them with a tailored step goal and then randomized them to one of four experimental conditions. The deposit contract required either a €10 fixed deposit or a customizable deposit with any amount between €1 and €20 upfront. Furthermore, the deposit was either not matched or 1:1 matched (doubled) with a reward provided by the experiment. During 20 intervention days, daily feedback on goal progress and incentive earnings was provided by the app. We investigated effects on the uptake (measured as agreeing to participate and paying the deposit) and effectiveness of behavioral adoption (measured as participant days goal achieved). Findings: Overall, the uptake of deposit contracts was 83.2%, and participants (n = 113) achieved 14.9 out of 20 daily step goals. A binary logistic regression showed that uptake odds were 4.08 times higher when a deposit was matched (p = .010) compared to when it was not matched. Furthermore, uptake odds were 3.53 times higher when a deposit was customizable (p = .022) compared to when it was fixed. Two-way ANCOVA showed that matching (p = .752) and customization (p = .143) did not impact intervention effectiveness. However, we did find a marginally significant interaction effect of deposit matching X deposit customization (p = .063, ηp2 = 0.032). Customization decreased effectiveness when deposits were not matched (p = .033, ηp2 = 0.089), but had no effect when deposits were matched (p = .776, ηp2 = 0.001). Conclusions: We provide the first experimental evidence that both matching and customization increase the uptake of a deposit contract for physical activity. We recommend considering both matching and customization to overcome lack of uptake, with a preference for customization since matching a deposit imposes significant additional costs. However, since we found indications that customizable deposits might reduce effectiveness (when the deposits are not matched), we urge for more research on the effectiveness of customizable deposit contracts. Finally, future research should investigate which participant characteristics are predictive of deposit contract uptake and effectiveness. Pre-registration: OSF Registries, https://osf.io/cgq48.Design AestheticsApplied Ergonomics and Desig
A shared PV system for transportation and residential loads to reduce curtailment and the need for storage systems
No full textThis paper proposes a shared multi-stakeholder PV system for traction substations and nearby residential loads to reduce the need for storage, AC grid exchange, and curtailment. The residential stakeholders offer both the base electrical load and the solar panels installation space needed by the traction stakeholder, who brings the peak load and investments to the former. Two case studies were conducted for one year in the city of Arnhem, The cy=Netherlands, using comprehensive and verified simulation models: A high-traffic and a low-traffic substation. The results showed a positive, synergetic benefit in reducing the PV system's excess energy and size requirement for any type of traction substations connected to any number of households. In one detailed example, the multi-stakeholder system suggested in this paper is shown to reduce curtailment by up to 80% in moments of zero-traction load. Generally, the direct load coverage of a PV system is increased by as much as 7 absolute percentage points to the single-stakeholder system when looking at energy-neutral system sizes. This multi-stakeholders system offers then an increase in the techno-economic feasibility of PV system integration in urban loads.DC systems, Energy conversion & Storag
Surface mass balance and climate of the Last Glacial Maximum Northern Hemisphere ice sheets: simulations with CESM2.1
No full textThe Last Glacial Maximum (LGM, from ∼26 to 20 ka BP) was the most recent period with large ice sheets in Eurasia and North America. At that time, global temperatures were 5–7 ∘C lower than today, and sea level ∼125 m lower. LGM simulations are useful to understand earth system dynamics, including climate–ice sheet interactions, and to evaluate and improve the models representing those dynamics. Here, we present two simulations of the Northern Hemisphere ice sheet climate and surface mass balance (SMB) with the Community Earth System Model v2.1 (CESM2.1) using the Community Atmosphere Model v5 (CAM5) with prescribed ice sheets for two time periods that bracket the LGM period: 26 and 21 ka BP. CESM2.1 includes an explicit simulation of snow/firn compaction, albedo, refreezing, and direct coupling of the ice sheet surface energy fluxes with the atmosphere. The simulated mean snow accumulation is lowest for the Greenland and Barents–Kara Sea ice sheets (GrIS, BKIS) and highest for British and Irish (BIIS) and Icelandic (IcIS) ice sheets. Melt rates are negligible for the dry BKIS and GrIS, and relatively large for the BIIS, North American ice sheet complex (NAISC; i.e. Laurentide, Cordilleran, and Innuitian), Scandinavian ice sheet (SIS), and IcIS, and are reduced by almost a third in the colder (lower temperature) 26 ka BP climate compared with 21 ka BP. The SMB is positive for the GrIS, BKIS, SIS, and IcIS during the LGM (26 and 21 ka BP) and negative for the NAISC and BIIS. Relatively wide ablation areas are simulated along the southern (terrestrial), Pacific and Atlantic margins of the NAISC, across the majority of the BIIS, and along the terrestrial southern margin of the SIS. The integrated SMB substantially increases for the NAISC and BIIS in the 26 ka BP climate, but it does not reverse the negative sign. Summer incoming surface solar radiation is largest over the high interior of the NAISC and GrIS, and minimum over the BIIS and southern margin of NAISC. Summer net radiation is maximum over the ablation areas and minimum where the albedo is highest, namely in the interior of the GrIS, northern NAISC, and all of the BKIS. Summer sensible and latent heat fluxes are highest over the ablation areas, positively contributing to melt energy. Refreezing is largest along the equilibrium line altitude for all ice sheets and prevents 40 %–50 % of meltwater entering the ocean. The large simulated melt for the NAISC suggests potential biases in the climate simulation, ice sheet reconstruction, and/or highly non-equilibrated climate and ice sheet at the LGM time.Physical and Space Geodes
Inhibitory effects of long chain fatty acids on anaerobic sludge treatment: Biomass adaptation and microbial community assessment
No full textThe study investigated the effects of long-chain fatty acids (LCFA) on anaerobic sludge treating lipid-rich wastewater. It involved batch experiments with three sludge samples: two acclimated to lipids and one non-acclimated. The experiments aimed to observe the degradation of LCFA, specifically oleate and palmitate, by dosing them at concentrations ranging from 50 to 600 mg/L. Measurements of the cumulative methane production and the LCFA concentration, quantified as fat, oil, and grease (FOG) were performed. To ensure the sludge was free from other biodegradable substrates, part of the samples was pre-incubated without feed. The tests were conducted with both pre-incubated and non-incubated inoculum sludge. The findings revealed that oleate was degraded more efficiently than palmitate across all sludge samples, with a greater conversion rate to methane. Sludge samples acclimated to lipids showed a superior capacity to degrade LCFA compared to non-acclimated ones. It was noted that at concentrations above 400 mg/L, the conversion of LCFAs to intermediate compounds was inhibited, although this did not affect the subsequent methane production. The study concludes with a recommendation for sludge adaptation strategies to boost the efficiency of anaerobic wastewater treatment systems dealing with lipid-rich waste. The presence of LCFA-degrading bacteria families like Kosmotogaceae, Petrotogaceae, and Synergistaceae in the acclimated sludge samples underscores the adaptation and potential for improved degradation performance.Sanitary EngineeringBT/Environmental Biotechnolog
Ambient air pollution and consumer spending: Evidence from Spain
No full textResearch on the economic burden of air pollution has focused primarily on its macroeconomic impact. However, as some studies have found that air pollution can lead to avoidance behavior–for example, reducing the time spent outdoors–we hypothesize that it can also influence consumer spending activity. We combine high frequency data on ozone and fine particulate pollution with daily consumer spending in brick-and-mortar retail in 129 postal codes in Spain during 2014 to estimate the association between the two. Using a linear fixed effects model, we find that a 1-standard deviation increase in ozone concentration (20.97 μg/m3) is associated with 3.9 percent decrease in consumer spending (95% CI: -0.066, -0.012; p<0.01). The association of fine particulate matter with consumer spending is, however, not statistically significant (β: 0.005; 95% CI: -0.009, 0.018; p>0.10). Further, we do not observe a sufficiently strong bounce-back in consumer spending in the day–or even the week–following higher ozone concentration. Also, we find that the relationship between ozone concentration and consumer spending is heterogeneous, with those aged below 25 and those aged 45 or above exhibiting stronger negative association. This research informs policymakers about a plausibly unaccounted cost of ambient air pollution, even at concentrations lower than the WHO air quality guideline for short-term exposure.Organisation & Governanc
Data Governance Challenges at Dutch Financial Services Firms
No full textIn 2006 already, Clive Humby said: “Data is the New Oil!” and like oil, data needs “infrastructure” to be gathered, analyzed and used. This infrastructure is called data governance and it is essential in today’s data-driven era to ensure availability, quality and security of an organization’s data. This is especially true for financial services firms, which deal with massive amounts of highly-sensitive personal data, such as names, dates of birth and bank details, and operate in a highly regulated environment. Therefore, it is essential that any new data governance policies, such as a transition to a cloud-based data governance policy, are implemented as quickly and efficiently as possible. Research thus far has primarily focused on the importance of data governance and developing data governance models. However, the implementation of data governance proves to be not without its barriers. Some research has been done into what barriers organizations encounter when attempting to implement new data governance policies, but strategies to deal with these barriers have not been found in existing literature. Therefore, this research seeks to answer the question:“How do managers at financial services firms in the Netherlands deal with the barriers to successfully implement new data governance?”This thesis used a literature study, twelve individual interviews with PwC employees who were heavily involved in data governance implementation processes at financial services firms in the Netherlands and a focus group interview with experts from PwC to determine what barriers financial services firms face when they are implementing new data governance, which strategies they use to deal with these barriers and what key factors influence the decision-making in this implementation process. These three elements were then used to find the answer to how managers at financial services firms in the Netherlands successfully implement new data governance.The research attempts to close the gap in the literature surrounding the general strategies that are used to navigate the barriers that inhibit (new) data governance implementation. Furthermore, it can help further identify which barriers (financial services) firms face when attempting to implement new data governance and aid in the development of more effective data governance framework. Additionally, the improved understanding of how financial services firms navigate the barriers that inhibit data governance implementation can help maintain trust in financial services firms and the financial system as a whole and it can aid in the development of more effective regulatory frameworks to increase how fast financial services firms are able to comply to them.The barriers financial firms face to implementing data governance that were found in this thesis were sorted into four broad categories: “Organizational culture/structure”, ”Senior management priority”, ”IT performance” and ”Lack of information”. Examples of these barriers are: a “restrictive mindset”, “unfocused strategy”, “incompatible IT systems” or a “lack of information on technoliiiogy”. The strategies the firms used to deal with these barriers were also sorted into four different categories: “Senior management vision/championing”, “Technological tools/skills”, “Stakeholder involvement/consensus” and the “Business case” strategy. Examples of these strategies are: “developing a global vision”, “standardization of technology”, “stakeholder involvement” and “building a broad business case”. A complete overview of the barriers and their corresponding strategies that were found in this thesis can be found in figure 4.1...Management of Technology (MoT