Enhancing autonomous systems with bayesian neural networks: a probabilistic framework for navigation and decision-making

SummaryWhile the choice of prior is one of the most critical parts of the Bayesian inference workflow, recent Bayesian deep learning models have often fallen back on vague priors, such as standard Gaussians. In this review, we highlight the importance of prior choices for Bayesian deep learning and present an overview of different priors that have been proposed for (deep) Gaussian processes, variational autoencoders and Bayesian neural networks. We also outline different methods of learning priors for these models from data. We hope to motivate practitioners in Bayesian deep learning to think more carefully about the prior specification for their models and to provide them with some inspiration in this regard.

The intermittence and fluctuation of renewable energy bring significant uncertainty to the power system, which enormously increases the operational risks of the power system. The development of efficient interval prediction models can provide data support for decision making and help improve the economy and reliability of energy interconnection operation. The performance of Bayesian deep learning models and Bayesian shallow neural networks in short-term interval prediction of photovoltaic power is compared in this study. Specifically, an LSTM Approximate Bayesian Neural Network model (ABNN-I) is built on the basis of the deep learning and Monte Carlo Dropout method. Meanwhile, a Feedforward Bayesian Neural Network (ABNN-II) model is introduced by Feedforward Neural Network and the Markov Chain Monte Carlo method. To better compare and verify the interval prediction capability of the ABNN models, a novel clustering method with three-dimensional features which include the number of peaks and valleys, the average power value, and the non-stationary measurement coefficient is proposed for generating sunny and non-sunny clustering sets, respectively. Results show that the ABNN-I model has an excellent performance in the field of photovoltaic short-term interval forecasting. At a 95% confidence level, the interval coverage from ABNN-I to ABNN-II can be increased by up to 3.1% and the average width of the interval can be reduced by 56%. Therefore, with the help of the high computational capacity of deep learning and the inherent ability to quantify uncertainty of the interval forecast from Bayesian methods, this research provides high-quality interval prediction results for photovoltaic power prediction and solves the problem of difficult modeling for over-fitting that exists in the training process, especially on the non-sunny clustering sets.

The emerging machine learning techniques provide great opportunities for optimal operation of chemical systems. This paper presents a Bayesian deep reinforcement learning method for the optimization of a fluid catalytic cracking (FCC) unit, which is a key process in the petroleum refining industry. Unlike the traditional reinforcement learning (RL) methods that use deterministic network weights, Bayesian neural networks are incorporated to represent the RL agent. The Bayesian treatment is integrated with the primal-dual method to handle the process constraints. Simulated experiments for FCC determined that the proposed algorithm achieves more stable control performance and higher economic profits, especially under parameter fluctuations and external disturbances.

In this paper, we propose a Bayesian Deep Learning (BDL) framework to model uncertainty and predict the performance of terahertz (THz) biosensors with a graphene and molybdenum disulfide (MoS2) coating for AML biomarker detection. Although there have been studies on the individual advantage of these 2D materials for biosensing, a comparative analysis taking into account predictive uncertainty is still insufficient. To this end, we have generated a high-fidelity simulation dataset from full-wave EM simulations of DSSRR structures over the 0.1–2.5 THz frequency range. Realistic geometrical and dielectric modifications have been incorporated to mimic bio-sensing conditions. An approach based on a Bayesian Neural Network (BNN) with Monte Carlo dropout was employed for predicting sensitivity, Q-factor, resonance shift, and absorption, along with the estimation of aleatoric, as well as epistemic, uncertainty. Our results demonstrated a trade-off between material types: MoS2 sensors showed higher sensitivity (3548 GHz/RIU) but with a larger prediction uncertainty range of ±118 GHz/RIU; on the other hand, graphene-based sensors exhibited a better spectral resolution (Q = 48.5) and a more reliable QV prediction range of ±42 GHz/RIU. The uncertainty study further revealed that graphene demonstrated a predominance for aleatoric uncertainty (68%), classifying them as predictable physical characteristics, while MoS2 presents a higher epistemic one (55%), indicating sensitivity towards underrepresented design cases. We present a material selection algorithm based on utility that balances sensitivity, resolution, and uncertainty, demonstrating that MoS2 is the best choice for early screening, while graphene is more suitable for high-precision diagnostics. This study offers a scalable and reliable AI framework for quick, uncertainty-aware optimization of THz biosensors, which is directly applicable to clinical diagnostics and 2D-material-based photonic design.

With the availability of data and computational technologies in the modern world, machine learning (ML) has emerged as a preferred methodology for data analysis and prediction. While ML holds great promise, the results from such models are not fully unreliable due to the challenges introduced by uncertainty. An ML model generates an optimal solution based on its training data. However, if the uncertainty in the data and the model parameters are not considered, such optimal solutions have a high risk of failure in actual world deployment. This paper surveys the different approaches used in ML to quantify uncertainty. The paper also exhibits the implications of quantifying uncertainty when using ML by performing two case studies with space physics in focus. The first case study consists of the classification of auroral images in predefined labels. In the second case study, the horizontal component of the perturbed magnetic field measured at the Earth’s surface was predicted for the study of Geomagnetically Induced Currents (GICs) by training the model using time series data. In both cases, a Bayesian Neural Network (BNN) was trained to generate predictions, along with epistemic and aleatoric uncertainties. Finally, the pros and cons of both Gaussian Process Regression (GPR) models and Bayesian Deep Learning (DL) are weighed. The paper also provides recommendations for the models that need exploration, focusing on space weather prediction.

The Bayesian network (BN) and influence diagram (ID) are used to develop a framework for post-earthquake risk assessment and decision making for infrastructure systems. The BN is used to model the earthquake hazard and component and system performance, and to update these models probabilistically in light of information gained from ground motion or structural health-monitoring sensors, or from observation of system or component states. The BN is extended by addition of decision and utility nodes to construct an ID, which is used for post-earthquake decision making regarding the type of inspection to perform or for setting the operational levels of components. A value-of-information heuristic is proposed to determine the optimal temporal sequence of component inspections. The methodology is demonstrated by its application to a hypothetical model of a segment of the proposed California high-speed rail system. Although the focus in this paper is on the earthquake hazard, the methodology can be used for infrastructure risk assessment and decision making for many other kinds of hazards. The main contribution of the paper lies in the integration of BN and ID models for the hazard and the infrastructure system and its components in a logical framework that allows rapid decision making to mitigate infrastructure losses after a major hazard.

Probabilistic remaining useful life prediction without lifetime labels: A Bayesian deep learning and stochastic process fusion method

CyberPhysical Systems (CPS) are increasingly targeted by sophisticated cyber threats, necessitating advanced Intrusion Detection Systems (IDS) capable of probabilistic risk assessment and adaptive defense mechanisms. Traditional deep learningbased IDS models, while effective in pattern recognition, lack uncertainty quantification, leading to unreliable classifications in dynamic attack environments. Bayesian Deep Learning (BDL)offers a principled approach to addressing this challenge by integrating probabilistic inference for enhanced threat detection and risk estimation. This book chapter explores the role of Bayesian Neural Networks (BNNs) in IDS, emphasizing their ability to model epistemic and aleatoric uncertainties, thereby improving detection accuracy and decision confidence. the chapter presents Bayesian Hyperparameter Optimization (BHO) as a scalable solution to optimize IDS model parameters, ensuring efficient and adaptive learning against evolving attack patterns. The integration of Bayesian techniques in both anomalybased and signaturebased IDS frameworks was examined, highlighting their potential in reducing false positives and mitigating adversarial attacks., case studies and empirical evaluations are provided to demonstrate the realworld applicability of BDLdriven IDS in securing industrial control systems, smart grids, and IoT networks. The findings underscore the importance of uncertaintyaware AIdriven cybersecurity frameworks for robust and resilient CPS protection.

The advancement of Maritime Autonomous Surface Ships (MASS) relies heavily on data-driven knowledge discovery and situational awareness to ensure safe and efficient navigation. This PhD by Published Works comprises five peer-reviewed journal papers, contributing to the integration of Artificial Intelligence (AI), machine learning, deep learning, and Bayesian Network (BN) into key areas of maritime safety and security. The research aims to enhance risk assessment, piracy incident analysis, ship trajectory prediction, and autonomous route planning, forming a cohesive framework for intelligent decision-making in MASS operations. The first contribution focuses on data-driven risk analysis of global maritime accidents, employing a data-driven BN model to quantify accident risks, identify critical factors, and enhance safety management strategies. The second contribution addresses spatio-temporal piracy pattern analysis, utilising AI-driven methodologies to detect high-risk zones and extract patterns of piracy incidents, thereby improving security measures and navigational decision-making for MASS. The third and fourth contributions focus on ship trajectory prediction and route planning, which are crucial for autonomous navigation. A systematic review of machine learning and deep learning-based trajectory prediction methods highlights the effectiveness of data-driven models over traditional approaches. Further, an unsupervised route planning model integrates Automatic Identification System (AIS) data and machine learning techniques, enabling autonomous vessels to adapt dynamically to changing maritime conditions. The final contribution refines AIS-based trajectory prediction frameworks, leveraging machine learning methodologies to improve predictive accuracy and enhance collision avoidance capabilities. By synthesising insights from these studies, this research establishes a holistic framework for data-driven knowledge discovery and situational awareness in MASS. The integration of this research content and outcomes ensures that autonomous vessels can navigate safely, mitigate operational risks, and enhance adaptability to evolving maritime conditions. This study advances AI-driven autonomous navigation by bridging data-driven knowledge discovery with real-time maritime decision-making, providing a resilient and adaptive foundation for next-generation MASS technology.

With the advent of increased computational resources and improved algorithms, machine learning-based models are being increasingly applied to complex problems in particle accelerators. However, such data-driven models may provide overly confident predictions with unknown errors and uncertainties. For reliable deployment of machine learning models in high-regret and safety-critical systems such as particle accelerators, estimates of prediction uncertainty are needed along with accurate point predictions. In this investigation, we evaluate Bayesian neural networks (BNN) as an approach that can provide accurate predictions along with reliably quantified uncertainties for particle accelerator problems, and compare their performance with bootstrapped ensembles of neural networks. We select three accelerator setups for this evaluation: a storage ring, a photoinjector, and a linac. The problems span different data volumes and dimensionalities (e.g., scalar predictions as well as image outputs). It is found that BNN provide accurate predictions of the mean along with reliable estimates of predictive uncertainty across the test cases. In this vein, BNN may offer an attractive alternative to deterministic deep learning tools to generate accurate predictions with quantified uncertainties in particle accelerator applications.

Estimating the contribution of forests to carbon sequestration is commonly done by applying forest growth models. Such models inherently use field observations, such as leaf area index (LAI), whereas relevant information is also available from remotely sensed images. The purpose of this study is to improve the LAI estimated from the Physiological Principles Predicting Growth (3-PG) model by combining its output with LAI derived from Advanced Spaceborne Thermal Emission and Reflection Radiometer (ASTER) satellite imagery. A Bayesian network (BN) approach is proposed to take care of the different structure of the inaccuracies in the two data sources. It addresses the bias in the 3-PG model and the noise of the ASTER images. Moreover, the EM algorithm is introduced into BN to estimate missing the LAI ASTER data, since they are not available for long time series due to the atmospheric conditions. This paper shows that the outputs obtained with the BN were more accurate than the 3-PG estimate, as the root mean square error reduces to 0.46, and the relative error to 5.86 percent. We conclude that the EM-algorithm within a BN can adequately handle missing LAI ASTER values, and BNs can improve the estimation of LAI values. Ultimately, this method may be used as a predicting model of LAI values, and handling the missing data of ASTER images time series.

We propose a Bayesian deep learning framework for model driven online sparse channel estimation task in Multi-user MIMO systems. Tools from Bayesian neural network and stochastic variational Bayesian Inference are utilized to capture aleatoric and epistemic uncertainty estimates. We treat the network prediction as an auxiliary variable to allow inference performance to be unaffected by the stage of training of the network. In addition to providing uncertainty estimates, being Bayesian, the framework enables us the possibility to marginalize over penalty parameters and is well suited for online scenario with changing environments. Our simulations show that the framework is robust to model mismatch, and efficiently captures uncertainty in the predictions. © 1983-2012 IEEE.

Unmanned aerial vehicles (UAVs), also known as drones, have gained considerable interest among academics in recent years, which significantly increases the demand for autonomous navigation systems for UAVs. In this study, autonomous navigation is addressed as a two-stage problem. The first stage is the obstacle detection phase, during which the UAVs leverage their onboard sensors to perceive the surroundings and detect obstacles. The second stage is the obstacle avoidance phase, where the UAVs make decisions to continue operations without colliding with anything while approaching the target. This study proposes a self-supervised learning-based depth prediction framework as the criterion for obstacle proximity detection. The framework consists of two subnetworks: the depth and pose estimation modules. The former predicts depth values, while the latter evaluates the object’s translation and rotation matrix. The framework leverages the subnetworks’ outputs to reconstruct the projection relationships of pixels between adjacent frames to optimize the training process. To address the challenge of UAVs operating in large-scale environments, a modified loss function that combines photometric loss and edge-aware smoothness is presented to increase the depth prediction module’s accuracy. This study also presents a deep reinforcement learning-based obstacle avoidance system that takes depth maps as inputs to generate evasive maneuvers. The system works independently from priori maps and models while it learns the collision-free trajectory from interacting with the environment. This study deploys value-based algorithms to learn the nonlinear mapping between the continuous inputs and discrete output motion commands. Furthermore, the proposed system is validated on different simulation scenarios. It is capable of completing the obstacle avoidance task with relatively optimal trajectories in varied environments, demonstrating its superior effectiveness and transferability. Keywords: unmanned aerial vehicles, urban airspace, obstacle detection and avoidance, self-supervised learning, deep reinforcement learning.

Uncertainty quantification for appliance recognition in non-intrusive load monitoring using Bayesian deep learning

<strong class="journal-contentHeaderColor">Abstract.</strong> Deep learning techniques have been increasingly used in flood management to overcome the limitations of accurate, yet slow, numerical models and to improve the results of traditional methods for flood mapping. In this paper, we review 58 recent publications to outline the state of the art of the field, identify knowledge gaps, and propose future research directions. The review focuses on the type of deep learning models used for various flood mapping applications, the flood types considered, the spatial scale of the studied events, and the data used for model development. The results show that models based on convolutional layers are usually more accurate, as they leverage inductive biases to better process the spatial characteristics of the flooding events. Models based on fully connected layers, instead, provide accurate results when coupled with other statistical models. Deep learning models showed increased accuracy when compared to traditional approaches and increased speed when compared to numerical methods. While there exist several applications in flood susceptibility, inundation, and hazard mapping, more work is needed to understand how deep learning can assist in real-time flood warning during an emergency and how it can be employed to estimate flood risk. A major challenge lies in developing deep learning models that can generalize to unseen case studies. Furthermore, all reviewed models and their outputs are deterministic, with limited considerations for uncertainties in outcomes and probabilistic predictions. The authors argue that these identified gaps can be addressed by exploiting recent fundamental advancements in deep learning or by taking inspiration from developments in other applied areas. Models based on graph neural networks and neural operators can work with arbitrarily structured data and thus should be capable of generalizing across different case studies and could account for complex interactions with the natural and built environment. Physics-based deep learning can be used to preserve the underlying physical equations resulting in more reliable speed-up alternatives for numerical models. Similarly, probabilistic models can be built by resorting to deep Gaussian processes or Bayesian neural networks.