A Multiple-Scale Model of Wind-Prediction Uncertainty and Application to Trajectory Prediction

Uncertainty in predictions of oil spill trajectories in a coastal zone

Brain tissue and tumor extraction plays a pivotal role in medical care and clinical research. Leveraging the diverse and complementary information provided by different imaging modalities is crucial to the success of these applications. However, existing deep learning-based methods exhibit limitations in two major aspects. First, they ignore the incorporation of task-oriented regularization when fusing multi-modal features, leading to suboptimal latent space learning. Second, these methods tend to overlook the explicit modeling and exploitation of prediction uncertainty, despite the strong correlation between prediction uncertainty and errors. To address these issues, we propose a novel lightweight network with uncertainty-guided latent space refinement for multi-modal brain tissue and tumor extraction, called UMNet. Particularly, UMNet features a modality-specific uncertainty-regularized feature fusion module (M-SUM), which facilitates latent space refinement to enable a more informed and effective aggregation of complementary multi-modal information. Additionally, we design an uncertainty-enhanced loss function (U-Loss) to explicitly harness the connection between prediction uncertainty and errors. Experimental results demonstrate that our proposed UMNet achieves promising performance, outperforming state-of-the-art methods for both brain tissue and tumor extraction tasks.

Recent deep learning methods for vessel trajectory prediction are able to learn complex maritime patterns from historical automatic identification system (AIS) data and accurately predict sequences of future vessel positions with a prediction horizon of several hours. However, in maritime surveillance applications, reliably quantifying the prediction uncertainty can be as important as obtaining high accuracy. This article extends deep learning frameworks for trajectory prediction tasks by exploring how recurrent encoder–decoder neural networks can be tasked not only to predict but also to yield a corresponding prediction uncertainty via Bayesian modeling of aleatoric and epistemic uncertainties. We compare the prediction performance of two different models based on labeled or unlabeled input data to highlight how uncertainty quantification and accuracy can be improved by using, if available, additional information on the intention of the ship (e.g., its planned destination).

Aiming at the problems of ID switching and tracking performance degradation caused by frequent occlusion and similar appearance of the tracked objects in dense scenes, a multi-object tracking method named TPFairMOT based on trajectory prediction and FairMOT is proposed in this paper. In the trajectory prediction branch, the object position of the future frame is predicted by using the object bounding box of the past frame and the velocity information learning network parameters, which overcomes the prediction failure caused by the uncertain motion state after the object is occluded in the tracking process. Secondly, the joint learning framework is used to combine the trajectory prediction branch with the detection and re-identification branch, and the tracking error caused by the high similarity between multiple objects in the tracking process is solved by integrating the appearance features and motion features of the tracked objects. Finally, MOTChallenge benchmarks (IDF1, IDs, MOTA, MT, and ML) are introduced to evaluate TPFairMOT, and different trajectory prediction strategies (FairMOT_KF and TPFairMOT_RNN) are used on FairMOT and TPFairMOT for comparative analysis. It is proved that the accuracy and ID switching times of trajectory prediction in this paper are better than other strategies. In addition, TPFairMOT, TPFairMOT_RNN, and FairMOT were compared on the public data sets MOT16, MOT17, and MOT20. The results show that TPFairMOT reduces the number of ID switching when the object is occluded, maintains the long-term validity of the identity information, and demonstrate good anti-occlusion performance.

Digital weld quality assurance systems are increasingly used to capture local geometrical variations that can be detrimental for the fatigue strength of welded components. In this study, a method is proposed to determine the required scanning sampling resolution for proper fatigue assessment. Based on FE analysis of laser-scanned welded joints, fatigue failure probabilities are computed using a Weakest-link fatigue model with experimentally determined parameters. By down-sampling of the scanning data in the FE simulations, it is shown that the uncertainty and error in the fatigue failure probability prediction increases with decreased sampling resolution. The required sampling resolution is thereafter determined by setting an allowable error in the predicted failure probability. A sampling resolution of 200 to 250 μm has been shown to be adequate for the fatigue-loaded welded joints investigated in the current study. The resolution requirements can be directly incorporated in production for continuous quality assurance of welded structures. The proposed probabilistic model used to derive the resolution requirement accurately captures the experimental fatigue strength distribution, with a correlation coefficient of 0.9 between model and experimental failure probabilities. This work therefore brings novelty by deriving sampling resolution requirements based on the influence of stochastic topographical variations on the fatigue strength distribution.

The 4D trajectory is a multi-dimensional time series with plentiful spatial-temporal features and has a high degree of complexity and uncertainty. Aiming at these features of aircraft flight trajectory and the problem that it is difficult for existing trajectory prediction methods to extract spatial-temporal features from the trajectory data at the same time, we propose a novel 4D trajectory prediction hybrid architecture based on deep learning, which combined Convolutional Neural Network (CNN) and Long Short-Term Memory (LSTM). An 1D convolution is used to extract the spatial dimension feature of the trajectory, and LSTM is used to mine the temporal dimension feature of the trajectory. Hence the high-precision prediction of the 4D trajectory is realized based on the sufficient fusion of the above features. We use real Automatic Dependent Surveillance -Broadcast (ADS-B) historical trajectory data for experiments and compare the proposed method with a single LSTM model and BP model on the same data set. The experimental results show that the trajectory prediction accuracy of the CNN-LSTM hybrid model is superior to a single model. The prediction error is reduced by an average of 21.62% compared to the LSTM model and by an average of 52.45% compared to the BP model. It provides a certain reference for the trajectory prediction research and Air Traffic Management decision-making.

The force tracking error is dependent on both the position tracking error and the estimated parameter error so that, in constrained adaptive robot control, the convergence of the estimated parameter error becomes more important than in unconstrained adaptive robot control. However, in the direct adaptive controller, the parameter adaptation is driven only by the tracking error in the joint motion, while in the indirect adaptive controller, the parameter estimation is driven only by the prediction error in the joint torque. Based on this observation a new combined direct and indirect adaptive controller driven by both the tracking error and the prediction error is developed for constrained robots with inertia parameter uncertainty. Stability of the combined adaptive controllers is established in the Lyapunov sense. The combined control improves both position and force tracking performance compared with direct adaptive control. The computer simulations are presented to demonstrate the proposed schemes.

Humans form sequences of event models-representations of the current situation-to predict how activity will unfold. Multiple mechanisms have been proposed for how the cognitive system determines when to segment the stream of behavior and switch from one active event model to another. Here, we constructed a computational model that learns knowledge about event classes (event schemas), by combining recurrent neural networks for short-term dynamics with Bayesian inference over event classes for event-to-event transitions. This architecture represents event schemas and uses them to construct a series of event models. This architecture was trained on one pass through 18 h of naturalistic human activities. Another 3.5 h of activities were used to test each variant for agreement with human segmentation and categorization. The architecture was able to learn to predict human activity, and it developed segmentation and categorization approaching human-like performance. We then compared two variants of this architecture designed to better emulate human event segmentation: one transitioned when the active event model produced high uncertainty in its prediction and the other transitioned when the active event model produced a large prediction error. The two variants learned to segment and categorize events, and the prediction uncertainty variant provided a somewhat closer match to human segmentation and categorization-despite being given no feedback about segmentation or categorization. These results suggest that event model transitioning based on prediction uncertainty or prediction error can reproduce two important features of human event comprehension.

The prediction accuracy of 4D (four-dimensional) trajectory is crucial for aviation safety and air traffic management. Firstly, the sine chaotic mapping is employed to enhance the sparrow search algorithm (Sine-SSA). This enhanced algorithm optimizes the threshold parameters of the BP (back propagation) neural network (Sine-SSA-BP), thereby improving the quality of the initial solution and enhancing global search capability. Secondly, the optimal weight thresholds obtained from the Sine-SSA algorithm are integrated into the BP neural network to boost its performance. Subsequently, the 4D trajectory data of the aircraft serve as input variables for the Sine-SSA-BP prediction model to conduct trajectory predictions. Finally, the prediction results from three models are compared against the actual aircraft trajectory. It is found that within the specified time series, the errors in longitude, latitude, and altitude for the Sine-SSA-BP prediction model are significantly smaller than those of the simple BP and SSA-BP models. This indicates that the Sine-SSA-BP model can achieve high-precision 4D trajectory prediction. The accuracy of trajectory prediction is notably improved by the sparrow search algorithm optimized with sine chaotic mapping, leading to faster convergence and better prediction outcomes, which better meet the requirements of aviation safety and control.

Adaptive learning requires interpreting prediction errors in light of environmental dynamics. However, environments may not only differ in terms of when changes occur, but also how they occur: some nonstationary processes in nature exhibit slow drifts over time, while others show more abrupt changes. This raises the question as to whether humans can adapt their learning to reflect the generative structure of their environment. Here, we compared how humans learn in two canonical nonstationary environments: abrupt change points versus gradual random walks. Using a predictive inference task and a unified Bayesian framework, we show across two behavioral experiments that humans adapt their learning normatively between these two environments. Notably, identical prediction errors were interpreted differently across them. Large prediction errors triggered sharp increases in learning rate under change-point but not random-walk dynamics, where small and large errors were more equally weighted. This matched the predictions of a normative Bayesian model, which itself adopted the appropriate generative model for each environment. In addition, we found that humans could dissociate two latent variables that needed to be jointly inferred: the mean and variance (or stochasticity). This was demonstrated by explicit uncertainty reports that closely tracked the current variance, and only showed sustained changes following change points in variance but not mean. Together, these results show that humans adapt their learning strategy to both how and what aspects of the environment are changing. They establish a unified computational account of adaptive learning across different environmental dynamics.

Sense of agency is the sense that one is the cause of an action and controls the events in the outside world. We propose a mathematical model that explains the sense of agency using information theoretic free energy. The model predicts that the prediction error and uncertainty interactively affect the sense of agency. In an experiment in which the prediction error and observation uncertainty were manipulated using response delay and number of sensory signals from different sensory modalities, we observed an interaction between response delay and number of sensory signals. The results support the model prediction suggesting that, when the prediction error is sufficiently small, uncertainty reduces the sense of agency; by contrast, when the prediction error is sufficiently large, uncertainty increases the sense of agency. These model-based findings provide a guide for optimizing operational response to prevent lack of sense of agency when the response delay varies.

The SynchronyTM Respiratory Tracking System (SRTS) component of the CyberKnife® Robotic Radiosurgery System (Accuray, Inc., Sunnyvale CA) enables real-time tracking of moving targets by modeling the correlation between the targets and external surrogate light-emitting diode (LED) markers placed on the patient's chest. Previous studies reported some cases with respiratory phase shifts between lung tumor and chest wall motions. In this study, the impacts of respiratory phase shifts on the motion-tracking accuracy of the SRTS were investigated. A plastic scintillator was used to detect the position of the x-ray beams. The scintillation light was recorded using a camera in a dark room. A moving phantom moved a U-shaped frame on the scintillator with a 4th power of sinusoidal functions. Three metallic markers for motion tracking and four fluorescent tapes were attached to the frame. The fluorescent tapes were used to identify phantom position and respiratory phase for each video frame. The beam positions collected, when considered relative to the phantom motion, represent the degree of tracking error. Beam position was calculated by adding error value to phantom position. Motions with respiratory phase shifts between the target and an extra stage mimicking chest wall motion were also tested for LED markers. Log files of the SRTS were analyzed to evaluate correlation errors. When target and LED marker motions were synchronized with a respiratory cycle of 4s, the maximum tracking errors for 90% and 95% of beam-on time were 1.0mm and 1.2mm, respectively. The frequency of tracking errors increased when LED marker motion phase preceded target motion. Tracking errors that corresponded to 90% beam-on time were within 2.4mm for 5-15% of phase shifts. In contrast, the tracking errors were very large when the LED marker delayed to the target motions; the maximum errors of 90% beam-on time were 3.0, 3.8, and 7.5mm for 5%, 10%, and 15% of phase shifts, respectively. The patterns of the tracking errors derived from the scintillation light were very similar to those of the correlation data of the SRTS derived from the log files, indicating that the tracking errors caused mainly due to the errors in modeling the correlation data. With long respiratory cycle of 6s, the tracking errors were significantly decreased; the maximum tracking errors for 95% beam-on time were 1.6mm and 2.2mm for early and delayed LED motion. We have investigated the motion-tracking accuracy of the CyberKnife SRTS for cases with the respiratory phase shift between the target and the LED marker. The maximum tracking errors for 90% probability were within 2.4mm when the target delays to the LED markers. When LED marker delays, however, very large tracking errors were observed. With a long respiratory cycle, the tracking errors were greatly improved to less than 2.2mm. Coaching slow breathing will be useful for accurate motion tracking radiotherapy.

Abstract. The approach of conditional nonlinear optimal perturbation (CNOP) was previously proposed to find the optimal initial perturbation (CNOP-I) in a given constraint. In this paper, we extend the CNOP approach to search for the optimal combined mode of initial perturbations and model parameter perturbations. This optimal combined mode, also named CNOP, has two special cases: one is CNOP-I that only links with initial perturbations and has the largest nonlinear evolution at a prediction time; while the other is merely related to the parameter perturbations and is called CNOP-P, which causes the largest departure from a given reference state at a prediction time. The CNOP approach allows us to explore not only the first kind of predictability related to initial errors, but also the second kind of predictability associated with model parameter errors, moreover, the predictability problems of the coexistence of initial errors and parameter errors. With the CNOP approach, we study the ENSO predictability by a theoretical ENSO model. The results demonstrate that the prediction errors caused by the CNOP errors are only slightly larger than those yielded by the CNOP-I errors and then the model parameter errors may play a minor role in producing significant uncertainties for ENSO predictions. Thus, it is clear that the CNOP errors and their resultant prediction errors illustrate the combined effect on predictability of initial errors and model parameter errors and can be used to explore the relative importance of initial errors and parameter errors in yielding considerable prediction errors, which helps identify the dominant source of the errors that cause prediction uncertainties. It is finally expected that more realistic models will be adopted to investigate this use of CNOP.

Extreme value theory provides effective approaches and methods for estimating return levels RL (with a typical return period >100 years) of extreme events. However, the lack of sufficiently representative observations to properly fit extreme value distributions (EVDs) is a recurring problem for any metocean engineer in situations where the number of observations is limited or of poor quality [1]. To overcome this problem, augmenting the set of observations with complementary information sources is an interesting option. In this paper, we address this problem by fitting EVDs to both observations and predictions from machine learning models using the approach developed by [2]. By design, however, the predictions of machine learning models are uncertain because they are learned from a limited number of training samples. We therefore propose to explicitly take this error into account when inferring the EVD parameters within an approximate Bayesian computation (ABC) scheme combined with the Wasserstein distance [3].The added value of this ML approach, which takes prediction uncertainty into account, is shown for cyclone-induced waves in Guadeloupe (Lesser Antilles) using a large database of extreme waves (representative of 1,000 years of cyclonic activity) that were numerically calculated within [4]. A random forest (RF) regression model is trained to link cyclone characteristics (radius, atmospheric pressure, distance to the eye of the hurricane) to significant wave height, and the quantile variant of the RF model is then used to model prediction error within the ABC scheme. Comparison with the 100-year and 500-year RL reference solutions (calculated using the complete database) shows that the ML-based approach results in low bias and high reliability of RL estimates as well as gain in computational efficiency, even when the sample size is reduced by a factor up to 10 and even when the RF prediction error remains moderate with cross-validation coefficient of determination of 70–75%. The benefit of integrating the ML prediction error is shown in different contexts, both along Guadeloupe coasts and in deep ocean environments.[1] Jonathan et al. (2021). Ocean Engineering, doi:10.1016/j.oceaneng.2020.107725[2] Rohmer et al. (2023). Ocean Modelling, doi:10.1016/j.ocemod.2023.102275[3] Bernton et al. (2019). Journal of the Royal Statistical Society Series B, https://doi.org/10.1111/rssb.12312[4] Interreg Carib-Coast program, https://www.carib-coast.com/en/

Longitudinal Guidance of Unmanned Aerial Vehicle Using Integral Sliding Mode Control