SN computer science最新文献

Reciprocal communication between road users is a vital element of road user interaction. Non-cooperative game theory is an effective framework for modelling and characterising communicative behaviour between road users, which enables the study of emergent benefits for both the issuer and recipient of communicative signals. In this paper, we introduce discretionary communication to gain an advantage over the other road user by masking one's intent if beneficial to do so. We conduct a series of experiments with simulated interactions and compare interaction outcomes where communication is mandatory against those where communication is discretionary. Our findings further support the premise that non-cooperative game theory is an effective paradigm for modelling and producing emergent behaviours which benefit the network. Moreover, we see that including a layer of discretionary communication reaps benefits in interaction outcome to the communicator. It also provides benefits in safety to all parties involved above and beyond the benefits seen from mandatory communication.

Datasets used in machine learning often contain sensitive information, including personally identifiable health and financial details. A common challenge faced by organizations and researchers is the risk of privacy breaches when using real-world data. Synthetic data can be used as an alternative to the real-world data. In existing synthetic data generation techniques, an encoder processes the real-world data to map it into a lower-dimensional latent space. Random sampling is then performed in this latent space. Subsequently, a decoder network is utilized to generate synthetic data from these sampled points in the latent space. Such approaches typically require generating a large number of synthetic samples to approximate the performance of real-world data, subsequently slowing down downstream machine learning tasks. Addressing this, we introduce a combinatorial approach to sampling the latent space, motivated by our empirical findings within this study that most model predictions are largely influenced by interactions between a few features. In some cases, just using a small number of features produces accuracy better than using entire features. Through this approach, we generate samples that utilize t-way interactions among the t latent dimensions out of n. Our experimental results indicate that our approach requires fewer samples than traditional random sampling to achieve comparable model performance for real-world data sets. We also show that when integrated with a differentially private mechanism, our approach incurs a smaller decline in model performance than existing random sampling approach.

Telecommunications networks are essential to the functioning of modern society, and large-scale disruptions in these networks can significantly impact societal operations. This paper presents Tram (Telecommunications Related AcciMap), an accident analysis method developed to support the analysis of accidents in telecommunication networks. Tram aimed to allow lessons learned from an accident to be used to mitigate and/or prevent new accidents, ultimately enhancing societal resilience. Initially, we studied the state-of-the-art in accident analysis methods, concluding that we should start from the AcciMap method since this is the most popular and appropriate method for our purposes. Tram has been developed through iterative cycles in which improvements have been proposed and validated; this paper focuses on the last iteration. Tram is a validated method that supports comprehensive analyses of telecom accidents. The latest version added support to the representation and mitigation of positive feedback loops during telecom and information technology systems breakdowns, and can help prioritise the recommendations derived from an analysis. This paper demonstrates that Tram is a suitable method to analyse and learn lessons from accidents in the telecom domain. Particularly, our findings with the application of Tram to real-life accidents indicate that dividing the analysis process by participant expertise can negatively impact the efficiency of the overall process, so the partitioning of the analysis process should be carefully considered.

Accurately identifying the positions of products in outdoor environments-such as warehouses or industrial yards-presents unique challenges due to variable lighting, weather conditions, and the lack of fixed infrastructure. This work presents a vision-based drone system for product localization using QR code detection and relative positioning. The proposed system enables a UAV to autonomously scan an area, extract QR codes from captured video frames, and compute the spatial relationships between products using a trust-ability graph. Unlike traditional GPS- or RFID-based methods, our approach does not rely on external infrastructure, making it scalable and adaptable for outdoor and semi-structured environments. We demonstrate that the proposed algorithm achieves over 94% positioning accuracy in indoor settings and 80% in outdoor environments, even under occlusion and varying illumination. The key contributions of this work include: (1) a novel infrastructure-free method for product positioning based on relative spatial relationships, (2) the integration of trust-ability scoring to improve the reliability of detected positions, and (3) an extensive evaluation in real-world indoor and outdoor industrial scenarios. These results validate the potential of UAV-assisted inventory systems to enhance automation in logistics and warehouse management.

We have developed a framework for efficient privacy preserving multi-party querying (PPMQ) over federated graph databases, leveraging Secure Multi-Party Computation (SMPC) protocols to enhance data security. The system offers two distinct security protocols: a client-based protocol and a server-based protocol. In the client-based protocol, standard SMPC techniques are employed, allowing computations to be performed on data without exposing the data itself. The server-based protocol employs SMPC to facilitate secure data processing and is further enhanced by encrypted hashing, which adds an additional layer of security to prevent data exposure. We conducted experiments comparing PPMQ with Neo4j Fabric and two previous systems, SMPQ and Conclave. The results indicate that PPMQ's execution times and overheads are comparable to those of Neo4j Fabric, while outperforming both SMPQ and Conclave, demonstrating its superior efficiency. Additionally, PPMQ, like SMPQ and Conclave, utilises an honest but curious security model. However, it enhances the security of the server protocol, making it more robust against brute force attacks and providing stronger privacy guarantees than previous solutions.

This paper addresses the development of a model-based design approach to enhance the acceptance of safe manoeuvrability of autonomous vehicles (AVs) on highways. A variable trust control setting (TCS) is introduced that empowers users to 'feel in control' of the AV, potentially increasing confidence in, and acceptance of, the technology. This setting is grounded in deontological ethics and utilises virtual boundaries (VBs) to guide driving decisions, i.e., the distance between two AVs interacting with one another. The approach is simulated using a dynamic bicycle model that represents each AV, controlled through an adaptive model-predictive control (MPC) algorithm. The paper outlines the MPC approach, the dynamic bicycle model, and the associated velocity control algorithm. Metrics are introduced to quantify safety of specific AV manoeuvres during interactions with other AVs, enabling the examination of various scenarios. A novel simulation package has been developed to investigate the impact of the proposed variable TCS, focusing on how VBs and steering limitations influence the safety and comfort of AVs during overtaking manoeuvres. The findings demonstrate the effectiveness of this approach, showing that it could potentially allow users to actively manage the safety and comfort aspects of AV operation.

Despite the recent developments in VR, maintaining photorealism is difficult due to the increased bandwidth capabilities and computational resources required. To make VR more affordable, techniques such as Foveated Rendering (FR) offer promising ways to optimise rendering without compromising the user experience. Near-eye displays with 6DOF tracking enable users to freely move through the environment. This was previously impossible with traditional displays. This work aims to disentangle the effect of type of ego-movement (Active versus Implied) and task type (Simple Fixations versus a task involving Fixations, Discrimination, and Counting) on a dynamic FR method developed using Variable Rate Shading (VRS) (a quarter of the native shading rate is used in the visual periphery). We also explore if the aforementioned effects are consistent under different visual behaviours (visual search versus tracking). Results show that participants actively moving and performing complex tasks (during visual search) are less sensitive to degradation than in other conditions, with only 31.7% of the FOV required to be rendered using full sampling. Additionally, we provide evidence for how instances of visual pursuit might influence these results; in this case, only 29.3% of the FOV rendered using full sampling is tolerated by participants.

Point-cloud data have become pivotal for three-dimensional scene analysis, yet robust real-time detection of humans remains challenging due to data sparsity, irregular sampling, and occlusions. In this study, we present a feature-engineered pipeline that uses a Random Forest Classifier (RFC) for efficient people detection in high-resolution LiDAR point clouds. Our contributions include: (1) detailed parameterization of a ground-removal algorithm using region growing; a compact feature set of 15 geometric and intensity-based descriptors; (3) comprehensive evaluation metrics on two datasets; and (4) comparative analysis against MLP and PointNet baselines. Experiments demonstrate that our RFC achieves good results. These results validate the practical applicability of our approach for real-time, on-device human detection in point-cloud environments.

Quantitative analysis of risk models is essential to ensure the resilience of complex systems. Fault trees (FTs) form a ubiquitous prominent risk model, and unreliability is its key safety metric. As complex systems have larger and larger models, the complexity of algorithms computing unreliability is a pressing concern. Unfortunately, state-of-the-art algorithms, based on binary decision diagrams, do not give time complexity guarantees beyond a worst-case exponential bound. To address this issue, this paper introduces a new method to compute FT unreliability, extending the fast bottom-up algorithm for tree-shaped FTs to general FTs by framing its arithmetic in algebras of squarefree polynomials. We prove the validity of this algorithm, and that its time complexity is linear when the number of multiparent nodes is limited. Experiments establish the competitiveness of our new method.

Ensuring high-quality drinking water is a critical responsibility of water utilities, with chlorine being the main disinfectant typically used. Accurate estimation of chlorine concentrations in the dynamic environment of water distribution networks (WDNs) is essential to ensure safe water supply. This work introduces a comprehensive and carefully created benchmark for training and evaluation of chlorine concentration estimation methodologies in WDNs. The benchmark includes a diverse dataset of 18,000 scenarios of the widely studied 'Hanoi', 'Net1', and the more recent and complex 'CY-DBP' water networks, featuring various chlorine injection patterns to capture diverse physical dynamics. To provide baseline evaluations, we propose and evaluate two neural surrogate models for chlorine state estimation: a physics-informed Graph Neural Network (GNN) and a physics-guided Recurrent Neural Network (RNN).