Unknown Scenarios Research Articles

Unsupervised global-local domain adaptation with self-training for remote sensing image semantic segmentation

ABSTRACT Unsupervised domain adaptation (UDA) techniques have the potential to enhance the transferability of neural network models in unknown scenarios and reduce the labelling costs associated with unlabelled datasets. Popular solutions to this challenging UDA task are adversarial training and self-training. However, current adversarial-based UDA methods emphasize only global or local feature alignment, which is insufficient for tackling the domain shift. In addition, self-training-based methods inevitably produce many wrong pseudo labels on the target domain due to bias towards the source domain. To tackle the above problems, this paper proposes a hybrid training framework that integrates global-local adversarial training and self-training strategies to effectively tackle global-local domain shift. First, the adversarial approach measures the discrepancies between domains from domain and category-level perspectives. The adversarial network incorporates discriminators at the local-category and global-domain levels, thereby facilitating global-local feature alignment through multi-level adversarial training. Second, the self-training strategy is integrated to acquire domain-specific knowledge, effectively mitigating negative migration. By combining these two domain adaptation strategies, we present a more efficient approach for mitigating the domain gap. Finally, a self-labelling mechanism is introduced to directly explore the inherent distribution of pixels, allowing for the rectification of pseudo labels generated during the self-training stage. Compared to state-of-the-art UDA methods, the proposed method gains 3.2 % , 1.21 % , 5.86 % , 6.16 % mIoU improvements on Rural → Urban, Urban → Rural, Potsdam → Vaihingen, Vaihingen → Potsdam, respectively.

Active Labeling Aided Semi-Supervised Safety Assessment With Task-Related Unknown Scenarios

Active Labeling Aided Semi-Supervised Safety Assessment With Task-Related Unknown Scenarios

High-performance network attack detection in unknown scenarios based on improved vertical model

High-performance network attack detection in unknown scenarios based on improved vertical model

Performance of expert fuzzy systems for prediction of rabbit feed intake after weaning.

The postweaning phase is stressful for rabbits due to maternal separation and the introduction of solid feed. Stress can be aggravated when animals are subjected to thermal discomfort. Therefore, thermal variables, such as air temperature, which influence the productive and physiological performance of animals, require greater control in this phase of life of rabbits to ensure their well-being and productive efficiency. Thus, the objective of the present study was to develop and compare fuzzy inference systems (FISs) with different configurations to predict the feed intake (FI) of New Zealand White (NZW) rabbits subjected to different thermal conditions after weaning. The experiment lasted 14 days, and twelve rabbits between 30 and 43 days old were used. The animals were housed in air-conditioned wind tunnels and subjected to air temperatures of 20, 24, 28 and 32°C. For the FIS configurations, Mamdani inference with five defuzzification methods (center of gravity (COG), bisector of area (BOA), largest of maximum (LOM), mean of maximum (MOM) and smallest of maximum (SOM)) and Sugeno inference with two defuzzification methods (weighted average (WA) and weighted sum (WS)), were evaluated. In both inference methods, the input variables (air temperature and time after weaning) were represented by triangular, Gaussian or trapezoidal functions. In turn, the output variable (FI, g) was represented by triangular, Gaussian or trapezoidal functions in the Mamdani FIS and by singleton functions in the Sugeno FIS. Thus, all developed FISs were validated, and their results were compared to the experimental data using statistical indices. As a result, adequate FI prediction performances were obtained for rabbits using both inference methods, regardless of the configurations used in their development. However, the smallest simulation errors were obtained using the Sugeno FIS with Gaussian inputs and WA defuzzification and is therefore a system with greater generalization capacity for unknown scenarios. Thus, the developed models can be used as a support system for decisions on the management of rabbits, aiding the efficient production and welfare of the animals, as well as the maintenance of thermal variables through the activation of installed climate systems inside the rabbit production environment.Trial registration number: 085/17. Date of registration: 14/12/2017.

Path loss prediction for air-to-ground communication links via scenario transfer technology

Path loss (PL) is a significant channel parameter for the link budget in unmanned aerial vehicle-aided communications. This study introduces an innovative neural network model to estimate PL for air-to-ground communication links. Utilizing the geometric characteristics of varied physical environments, the model accurately predicts PL in diverse communication scenarios. A back-propagation neural network technique is introduced for extrapolating PL under both line-of-sight and non-line-of-sight conditions. A dataset acquisition strategy, comprising scenario reconstruction and advanced ray-tracing techniques, is employed to foster the model’s training and evaluation. Finally, the proposed model is fully trained in diverse communication scenarios, and then used to predict the PLs in a new communication scenario generated by the International Telecommunication Union standard at 28 GHz. The results demonstrate that the extrapolated PLs of the proposed model are well consistent with the reference results. As existing PL models and standard PL models aim at several specifically defined scenarios, the proposed model can predict the PLs in some undefined and unknown scenarios.

Consensus-Based Model Predictive Control for Active Power and Voltage Regulation in Active Distribution Networks

In this paper, a consensus-based model predictive control (Cb-MPC) scheme is proposed to control the active power and voltage at all nodes in grid-connected active distribution networks (ADNs) with multiple distributed energy resources (DERs). The proposed design methodology is based on a multiple-input multiple-output (MIMO) model of an ADN which accounts for both the internal and external interactions among the control loops of the DERs. To achieve the control objective, each DER unit is equipped with a controller–observer system. In particular, the observer implements the consensus algorithm to estimate the collective system state by exchanging data only with its neighbors. The scope of the controller is to solve the MPC optimal problem based on its collective state estimate, and, due to the presence of an integral term in the control action, it is robust against any unknown scenarios of the ADN, which are represented by uncertainty in the model parameters. The results of numerical simulations validate the effectiveness of the proposed method in the presence of unknown changes in the operating conditions of the ADN and of communication using a sample and hold function.

Mars Exploration: Research on Goal-Driven Hierarchical DQN Autonomous Scene Exploration Algorithm

In the non-deterministic, large-scale navigation environment under the Mars exploration mission, there is a large space for action and many environmental states. Traditional reinforcement learning algorithms that can only obtain rewards at target points and obstacles will encounter the problems of reward sparsity and dimension explosion, making the training speed too slow or even impossible. This work proposes a deep layered learning algorithm based on the goal-driven layered deep Q-network (GDH-DQN), which is more suitable for mobile robots to explore, navigate, and avoid obstacles without a map. The algorithm model is designed in two layers. The lower layer provides behavioral strategies to achieve short-term goals, and the upper layer provides selection strategies for multiple short-term goals. Use known position nodes as short-term goals to guide the mobile robot forward and achieve long-term obstacle avoidance goals. Hierarchical execution not only simplifies tasks but also effectively solves the problems of reward sparsity and dimensionality explosion. In addition, each layer of the algorithm integrates a Hindsight Experience Replay mechanism to improve performance, make full use of the goal-driven function of the node, and effectively avoid the possibility of misleading the agent by complex processes and reward function design blind spots. The agent adjusts the number of model layers according to the number of short-term goals, further improving the efficiency and adaptability of the algorithm. Experimental results show that, compared with the hierarchical DQN method, the navigation success rate of the GDH-DQN algorithm is significantly improved, and it is more suitable for unknown scenarios such as Mars exploration.

Expediting the Convergence of Global Localization of UAVs through Forward-Facing Camera Observation

In scenarios where the global navigation satellite system is unavailable, unmanned aerial vehicles (UAVs) can employ visual algorithms to process aerial images. These images are integrated with satellite maps and digital elevation models (DEMs) to achieve global localization. To address the localization challenge in unfamiliar areas devoid of prior data, an iterative computation-based localization framework is commonly used. This framework iteratively refines its calculations using multiple observations from a downward-facing camera to determine an accurate global location. To improve the rate of convergence for localization, we introduced an innovative observation model. We derived a terrain descriptor from the images captured by a forward-facing camera and integrated it as supplementary observation into a point-mass filter (PMF) framework to enhance the confidence of the observation likelihood distribution. Furthermore, within this framework, the methods for the truncation of the convolution kernel and that of the probability distribution were developed, thereby enhancing the computational efficiency and convergence rate, respectively. The performance of the algorithm was evaluated using real UAV flight sequences, a satellite map, and a DEM in an area measuring 7.7 km × 8 km. The results demonstrate that this method significantly accelerates the localization convergence during both takeoff and ascent phases as well as during cruise flight. Additionally, it increases localization accuracy and robustness in complex environments, such as areas with uneven terrain and ambiguous scenes. The method is applicable to the localization of UAVs in large-scale unknown scenarios, thereby enhancing the flight safety and mission execution capabilities of UAVs.

Integrable symplectic maps with a polygon tessellation

Identifying integrable dynamics remains a formidable challenge, and despite centuries of research, only a handful of examples are known to date. In this article, we explore a distinct form of area-preserving (symplectic) mappings derived from the stroboscopic Poincaré cross section of a kicked rotator—an oscillator subjected to an external force periodically switched on in short pulses. The significance of this class of problems extends to various applications in physics and mathematics, including particle accelerators, crystallography, and studies of chaos. Notably, Suris's theorem constrains the integrability within this category of mappings, outlining potential scenarios with analytic invariants of motion. In this paper, we challenge the assumption of the analyticity of the invariant by exploring piecewise linear transformations on a torus (T2) and associated systems on the plane (R2), incorporating arithmetic quasiperiodicity and discontinuities. Introducing a new automated technique, we discovered previously unknown scenarios featuring polygonal invariants that form perfect tessellations and, moreover, fibrations of the plane or torus. This work reveals a novel category of planar tilings characterized by discrete symmetries that emerge from the invertibility of transformations and are intrinsically linked to the presence of integrability. Our algorithm relies on the analysis of the Poincaré rotation number and its piecewise monotonic nature for integrable cases, contrasting with the noisy behavior in the case of chaos, thereby allowing for clear separation. Some of the newly discovered systems exhibit the peculiar behavior of “integrable diffusion,” characterized by infinite and quasirandom hopping between tiles while being confined to a set of invariant segments. Finally, through the implementation of a smoothening procedure, all mappings can be generalized to quasi-integrable scenarios with suppressed volume occupied by chaotic trajectories, thereby opening doors to potential practical applications. Published by the American Physical Society 2024

A framework of a data-driven model for ship performance

An accurate assessment of a ship's required power is increasingly relevant for ship operations. We used a simplified framework of a data-driven model to predict ship's fuel consumption. Our approach was based on the learning capabilities of a generalized AutoML (Automated Machine Learning) process, trained with a variety of databases obtained from Computational-Fluid-Dynamics (CFD) simulations or from simplified numerical methods. These CFD simulations were conducted by solving the Reynolds Averaged Navier-Stokes (RANS) equations, using the STARCCM + commercial CFD software to calculate the ship resistance at speed under different operating conditions. Initially, we conducted a statistical analysis to select the independent variables before fitting the regression models and identifying potentially wrong assumptions. The AutoML process allowed optimizing the model's hyperparameters and designing the topology of the neural networks. For a set of unknown scenarios, comparative predictions obtained from the data-driven model and from numerical simulations showed that the data-driven model, trained with results obtained from CFD simulations, accurately and efficiently predicted ship operational parameters under realistic operating conditions, thereby dispensing with the need to perform elaborate CFD computations. Specifically, this low-cost and efficient operational data-driven technique forecasted the ship's operational fuel consumption although only a limited amount of recorded operational data was available.

Evaluating the stealth of reinforcement learning-based cyber attacks against unknown scenarios using knowledge transfer techniques

Organizations are vulnerable to cyber attacks as they rely on computer networks and the internet for communication and data storage. While Reinforcement Learning (RL) is a widely used strategy to simulate and learn from these attacks, RL-guided offensives against unknown scenarios often lead to early exposure due to low stealth resulting from mistakes during the training phase. To address this issue, this work evaluates if the use of Knowledge Transfer Techniques (KTT), such as Transfer Learning and Imitation Learning, reduces the probability of early exposure by smoothing mistakes during training. This study developed a laboratory platform and a method to compare RL-based cyber attacks using KTT for unknown scenarios. The experiments simulated 2 unknown scenarios using 4 traditional RL algorithms and 4 KTT. In the results, although some algorithms using KTT obtained superior results, they were not so significant for stealth during the initial epochs of training. Nevertheless, experiments also revealed that throughout the entire learning cycle, Trust Region Policy Optimization (TRPO) is a promising algorithm for conducting cyber offensives based on Reinforcement Learning.

E-Watcher: insider threat monitoring and detection for enhanced security

Insider threats refer to harmful actions carried out by authorized users within an organization, posing the most damaging risks. The increasing number of these threats has revealed the inadequacy of traditional methods for detecting and mitigating insider threats. These existing approaches lack the ability to analyze activity-related information in detail, resulting in delayed detection of malicious intent. Additionally, current methods lack advancements in addressing noisy datasets or unknown scenarios, leading to under-fitting or over-fitting of the models. To address these, our paper presents a hybrid insider threat detection framework. We not only enhance prediction accuracy by incorporating a layer of statistical criteria on top of machine learning-based classification but also present optimal parameters to address over/under-fitting of models. We evaluate the performance of our framework using a real-life threat test dataset (CERT r4.2) and compare it to existing methods on the same dataset (Glasser and Lindauer 2013). Our initial evaluation demonstrates that our proposed framework achieves an accuracy of 98.48% in detecting insider threats, surpassing the performance of most of the existing methods. Additionally, our framework effectively handles potential bias and data imbalance issues that can arise in real-life scenarios.

Fuzzing Digital Twin With Graphical Visualization of Electronic AVs Provable Test for Consumer Safety

In electronic autonomous vehicles (AVs), provable and explainable safety becomes the critical protection for their consumers. While traditional safety test schemes can detect the unsafe factors of AVs, such existing schemes still leave a number of challenges especially for provable safety test of AVs. First, existing schemes cannot continually test all traffic scenarios in the time domain such as future unknown scenarios and the scenarios with the degraded performance of the AVs. Second, it is an open issue that quantifies safety and explains the relationships between safety testing and proof, especially how safe is enough and why tests can transform into a certain level under the safety proof scale. To address these challenges, we propose a fuzzing digital twin approach, DT-FT, to construct a provable safety scheme for AVs. Specifically, we propose a dynamic strategy to guarantee the safety of AVs in the time domain and design a coverage model to quantify the safety under the safety proof scale. Moreover, we propose an approximation theory for the safety of AVs based on formal proof. Finally, a graphical visualization-based provable safety test application case for consumers is shown and the simulation results demonstrate the feasibility and effectiveness of DT-FT.

Uncertainty-aware deep learning for reliable health monitoring in safety-critical energy systems

In recent years, significant advancements in deep learning technology have facilitated the development of intelligent health monitoring approaches for energy systems. However, when dealing with safety-critical energy systems, such as nuclear energy systems, conventional deep learning models with point estimation fail to account for the inherent uncertainty in the predictions. This limitation poses challenges for providing reliable and trustworthy decision support for critical operations. To overcome this challenge, this study proposes a novel intelligent monitoring approach that integrates uncertainty-aware deep neural networks. Firstly, a spatio-temporal state matrix-based signal preprocessing method is proposed to enhance feature extraction capabilities, enabling the effective integration of diverse multi-source data. Secondly, a probabilistic distribution is developed to generate predictive uncertainty for all network parameters, enabling the assessment of the confidence of the model’s outputs not only for known operation scenarios but also for unknown scenarios. Finally, the experiments are conducted using an established advanced nuclear energy research platform and a public nuclear accident simulation platform, ensuring the effectiveness and applicability of the proposed approach in practical settings. Overall, the proposed approach significantly enhances the reliability and trustworthiness of the monitoring outputs while mitigating the risks associated with the decision-making process in safety-critical energy systems.

Cointegration technique to account for environmental variability in a concrete dam

Within Operational Modal Analysis (OMA) and Structural Health Monitoring (SHM) frameworks, novelty analysis aims to detect structural anomalies by inspecting the time evolution of the modal parameters of a structure subjected to ambient (unknown) excitations. However, real structures are subjected not only to structural anomalies, but also to environmental and operational variability (EOV), so it is necessary to remove EOV effects from damage-sensitive features. The paper presents an application of Cointegration technique to account for EOV on natural frequencies: given a set of non-stationary time-series, Cointegration builds a stationary linear combination of them, the cointegration residual, which is purged from the common trends mainly associated with EOV. The coefficients of the linear combination are determined by a maximum-likelihood multivariate technique, employing the natural frequencies within a training period, in which the structure is supposed to be in healthy conditions, under normal EOV. Once trained, the computed coefficients are used to test natural frequencies from unknown scenarios. The obtained cointegration residual will maintain its stationarity, as long as the structure behaves in normal conditions. Compared to traditional techniques, the Cointegration relies on more general operations, such as the construction of a linear combination given a set of modal parameters. The application of the presented procedure is exemplified with data measured in Baixo Sabor dam (Portugal) during the first two years of monitoring, which include the first filling of the reservoir.

Vibration-based novelty detection using autoencoders: application to KW51 bridge

In recent years an increasing attention has been given to the application of Artificial Intelligence techniques within the context of Structural Health Monitoring (SHM) of Civil Engineering structures. Multiple advantages compared to traditional techniques based on Operational Modal Analysis (OMA) are achieved, such as a relatively low computational efforts and possible insensitivity to environmental and operational variability (EOV). The present paper proposes an autoencoder-based approach to automatically handle EOVs and, at the same time, to detect structural changes. The proposed procedure is based on the use of an autoencoder (AE) model to reconstruct the measurement data collected during continuous dynamic monitoring. In more details, an AE network is trained using the time series simultaneously collected by all the available channels during a reference period, in which the structure is supposed to be in healthy condition under normal EOV. Through the training procedure, the internal bottleneck layer of AE is supposed to learn how the variation of measured data is affected by the normal EOV. Subsequently, the trained AE is used to reconstruct the data collected from unknown scenarios, providing a reconstruction error (between the measured and the reconstructed signals) that increases as soon as the monitored system departs from its healthy condition. The application of the AE-based procedure is exemplified to the benchmark KW51 bridge (a steel bowstring railway bridge in Leuven, Belgium), showing that the AE network, trained simultaneously with all the available channels, is capable of detecting the structural changes due to a retrofitting performed on the bridge, under changing environmental and operational conditions.

Soft Hybrid Knowledge Distillation against deep neural networks

Traditional knowledge distillation approaches are typically designed for specific tasks, as they primarily distilling deep features from intermediate layers of a neural network, generally with ingeniously designed knowledge representations, which increase the difficulty of model development and interpretation. In contrast, we empirically show that a soft logits knowledge distillation is enough to significantly narrow the teacher–student model performance gap. In this paper, the Soft Hybrid Knowledge Distillation (SHKD) is proposed to generate representative features from the input and output location of models, and applicable to various teacher and student architectures. Specifically, the additional mixup task is first applied to the same batch of training images, which enriches the high-order feature information transferred by the model. And then, a top-k guided specification filter is added to the teacher model, to reduce the bias induced throughout the knowledge transfer process. Finally, the SHKD utilizes the same and different teacher–student model pairs of the deep neural network, which can be easily adjusted to the computational resources available in the current world. Extensive experiments are conducted to examine the efficiency, and show that the SHKD outperforms state-of-the-art knowledge distillation methods. With the SHKD, the high performance of models will be maintained, and the gap between the teacher model and the student model can be reduced, making the construction of deep neural networks more convenient to the unknown scenarios. The source code is available at https://github.com/lambett/SHKD.

Synthetic data generation for the continuous development and testing of autonomous construction machinery

Abstract The development and testing of autonomous systems require sufficient meaningful data. However, generating suitable scenario data is a challenging task. In particular, it raises the question of how to narrow down what kind of data should be considered meaningful. Autonomous systems are characterized by their ability to cope with uncertain situations, i.e. complex and unknown environmental conditions. Due to this openness, the definition of training and test scenarios cannot be easily specified. Not all relevant influences can be sufficiently specified with requirements in advance, especially for unknown scenarios and corner cases, and therefore the “right” data, balancing quality and efficiency, is hard to generate. This article discusses the challenges of automated generation of 3D scenario data. We present a training and testing loop that provides a way to generate synthetic camera and Lidar data using 3D simulated environments. Those can be automatically varied and modified to support a closed-loop system for deriving and generating datasets that can be used for continuous development and testing of autonomous systems.

Towards Multi-Scenario Power System Stability Analysis: An Unsupervised Transfer Learning Method Combining DGAT and Data Augmentation

With the expansion of power system simulation scale, intelligent stability analysis based on simulation data becomes more and more important. However, the changeable operating conditions and fault conditions will seriously affect the performance of the intelligent analysis model. This paper proposes an advanced unsupervised transfer learning (UTL) method combining dynamic graph attention network (DGAT) and data augmentation to identify the instability mode towards multiple known and unknown scenarios. For the pre-training stage, the DGAT model with data augmentation is used for instability mode identification based on the basic dataset. For the transfer learning stage, it starts when faced with new operating conditions or fault conditions different from the basic dataset. The UTL scheme takes the distribution differences among different categories of source domain and target domain as the loss to fine-tune the pre-trained model. Moreover, considering the difficulty of sample-labeling, the UTL scheme can make the pre-trained model adapt to them without additional labeled samples. Case studies are conducted on 8-machine 36-bus system and Northeast China Power Grid. The results verify the superiority of the DGAT model with data augmentation and confirm the UTL scheme can help the model adapt to multiple unknown scenarios.

How Does Traffic Environment Quantitatively Affect the Autonomous Driving Prediction?

Accurate trajectory prediction is essential for safe and efficient autonomous driving in complex traffic environments. While artificial intelligence has shown great promise in improving prediction accuracy, its inherent uncertainty and lack of explainability may lead to unpredictable failures, creating challenges for safety-critical decision-making. This study aims to address these challenges by exploring the impact of traffic environment on prediction algorithms. The study proposes a trajectory prediction framework with epistemic uncertainty estimation ability to output high uncertainty when facing unforeseeable or unknown scenarios. The framework analyzes the environmental effect on the trajectory prediction by considering scenario features and shifts. Features are divided into kinematic features of a target agent, features of surrounding traffic participants, and other scenario features. Feature correlation and importance analyses are performed to study their influence on prediction error and epistemic uncertainty. The impact of unavoidable distributional shifts in the real world on trajectory predictions is investigated using multiple intersection datasets. The results indicate that deep ensemble-based methods have advantages in improving robustness while estimating epistemic uncertainty. Consistent conclusions were obtained from the correlation and importance analyses, indicating that kinematic features of the target agent have relatively strong effects on both prediction error and epistemic uncertainty. Finally, the study analyzes the accuracy deterioration caused by distributional shifts and the potential of the deep ensemble-based method. Through deep ensemble, the errors of the prediction methods based on GRIP++ and Trajectron++ have been improved by 6.4% and 10.8% in the same-dataset test, and 6.3% and 10.8% in the cross-dataset test.