Showing 1–14 of 14 results for author: Mosleh, A

A Structured Review of Literature on Uncertainty in Machine Learning & Deep Learning

Authors: Fahimeh Fakour, Ali Mosleh, Ramin Ramezani

Abstract: The adaptation and use of Machine Learning (ML) in our daily lives has led to concerns in lack of transparency, privacy, reliability, among others. As a result, we are seeing research in niche areas such as interpretability, causality, bias and fairness, and reliability. In this survey paper, we focus on a critical concern for adaptation of ML in risk-sensitive applications, namely understanding a… ▽ More The adaptation and use of Machine Learning (ML) in our daily lives has led to concerns in lack of transparency, privacy, reliability, among others. As a result, we are seeing research in niche areas such as interpretability, causality, bias and fairness, and reliability. In this survey paper, we focus on a critical concern for adaptation of ML in risk-sensitive applications, namely understanding and quantifying uncertainty. Our paper approaches this topic in a structured way, providing a review of the literature in the various facets that uncertainty is enveloped in the ML process. We begin by defining uncertainty and its categories (e.g., aleatoric and epistemic), understanding sources of uncertainty (e.g., data and model), and how uncertainty can be assessed in terms of uncertainty quantification techniques (Ensembles, Bayesian Neural Networks, etc.). As part of our assessment and understanding of uncertainty in the ML realm, we cover metrics for uncertainty quantification for a single sample, dataset, and metrics for accuracy of the uncertainty estimation itself. This is followed by discussions on calibration (model and uncertainty), and decision making under uncertainty. Thus, we provide a more complete treatment of uncertainty: from the sources of uncertainty to the decision-making process. We have focused the review of uncertainty quantification methods on Deep Learning (DL), while providing the necessary background for uncertainty discussion within ML in general. Key contributions in this review are broadening the scope of uncertainty discussion, as well as an updated review of uncertainty quantification methods in DL. △ Less

Submitted 1 June, 2024; originally announced June 2024.

Proceedings of the 3rd International Workshop on Autonomous Systems Safety

Authors: Christoph Thieme, Marilia Ramos, Ingrid B. Utne, Ali Mosleh

Abstract: The International Workshop for Autonomous System Safety (IWASS) is a joint effort by the B. John Garrick Institute for the Risk Sciences at the University of California Los Angeles (UCLA-GIRS) and the Norwegian University of Science and Technology (NTNU). IWASS is an invitation-only event designed to be a platform for cross-industrial and interdisciplinary effort and knowledge exchange on autonomo… ▽ More The International Workshop for Autonomous System Safety (IWASS) is a joint effort by the B. John Garrick Institute for the Risk Sciences at the University of California Los Angeles (UCLA-GIRS) and the Norwegian University of Science and Technology (NTNU). IWASS is an invitation-only event designed to be a platform for cross-industrial and interdisciplinary effort and knowledge exchange on autonomous systems' Safety, Reliability, and Security (SRS). The workshop gathers experts from academia, regulatory agencies, and industry to discuss challenges and potential solutions for SRS of autonomous systems from different perspectives. It complements existing events organized around specific types of autonomous systems (e.g., cars, ships, aviation) or particular safety or security-related aspects of such systems (e.g., cyber risk, software reliability, etc.). IWASS distinguishes itself from these events by addressing these topics together and proposing solutions for SRS challenges common to different types of autonomous systems. IWASS 2022 was held on August 28th in Dublin, Ireland, and gathered 30 participants from 20 organizations from around the globe. In addition, a panel session at the European Safety and Reliability Conference (ESREL 2023) discussed the workshop's main conclusions and additional points with a larger audience. This report summarizes IWASS 2022 discussions. It provides an overview of the main points raised by a community of experts on the current status of autonomous systems SRS. It also outlines research directions for safer, more reliable and secure future autonomous systems. △ Less

Submitted 27 April, 2023; originally announced April 2023.

SeKron: A Decomposition Method Supporting Many Factorization Structures

Authors: Marawan Gamal Abdel Hameed, Ali Mosleh, Marzieh S. Tahaei, Vahid Partovi Nia

Abstract: While convolutional neural networks (CNNs) have become the de facto standard for most image processing and computer vision applications, their deployment on edge devices remains challenging. Tensor decomposition methods provide a means of compressing CNNs to meet the wide range of device constraints by imposing certain factorization structures on their convolution tensors. However, being limited t… ▽ More While convolutional neural networks (CNNs) have become the de facto standard for most image processing and computer vision applications, their deployment on edge devices remains challenging. Tensor decomposition methods provide a means of compressing CNNs to meet the wide range of device constraints by imposing certain factorization structures on their convolution tensors. However, being limited to the small set of factorization structures presented by state-of-the-art decomposition approaches can lead to sub-optimal performance. We propose SeKron, a novel tensor decomposition method that offers a wide variety of factorization structures, using sequences of Kronecker products. By recursively finding approximating Kronecker factors, we arrive at optimal decompositions for each of the factorization structures. We show that SeKron is a flexible decomposition that generalizes widely used methods, such as Tensor-Train (TT), Tensor-Ring (TR), Canonical Polyadic (CP) and Tucker decompositions. Crucially, we derive an efficient convolution projection algorithm shared by all SeKron structures, leading to seamless compression of CNN models. We validate SeKron for model compression on both high-level and low-level computer vision tasks and find that it outperforms state-of-the-art decomposition methods. △ Less

Submitted 12 October, 2022; originally announced October 2022.

A generic physics-informed neural network-based framework for reliability assessment of multi-state systems

Authors: Taotao Zhou, Xiaoge Zhang, Enrique Lopez Droguett, Ali Mosleh

Abstract: In this paper, we leverage the recent advances in physics-informed neural network (PINN) and develop a generic PINN-based framework to assess the reliability of multi-state systems (MSSs). The proposed methodology consists of two major steps. In the first step, we recast the reliability assessment of MSS as a machine learning problem using the framework of PINN. A feedforward neural network with t… ▽ More In this paper, we leverage the recent advances in physics-informed neural network (PINN) and develop a generic PINN-based framework to assess the reliability of multi-state systems (MSSs). The proposed methodology consists of two major steps. In the first step, we recast the reliability assessment of MSS as a machine learning problem using the framework of PINN. A feedforward neural network with two individual loss groups are constructed to encode the initial condition and state transitions governed by ordinary differential equations (ODEs) in MSS. Next, we tackle the problem of high imbalance in the magnitude of the back-propagated gradients in PINN from a multi-task learning perspective. Particularly, we treat each element in the loss function as an individual task, and adopt a gradient surgery approach named projecting conflicting gradients (PCGrad), where a task's gradient is projected onto the norm plane of any other task that has a conflicting gradient. The gradient projection operation significantly mitigates the detrimental effects caused by the gradient interference when training PINN, thus accelerating the convergence speed of PINN to high-precision solutions to MSS reliability assessment. With the proposed PINN-based framework, we investigate its applications for MSS reliability assessment in several different contexts in terms of time-independent or dependent state transitions and system scales varying from small to medium. The results demonstrate that the proposed PINN-based framework shows generic and remarkable performance in MSS reliability assessment, and the incorporation of PCGrad in PINN leads to substantial improvement in solution quality and convergence speed. △ Less

Submitted 30 November, 2021; originally announced December 2021.

An Uncertainty-Informed Framework for Trustworthy Fault Diagnosis in Safety-Critical Applications

Authors: Taotao Zhou, Enrique Lopez Droguett, Ali Mosleh, Felix T. S. Chan

Abstract: There has been a growing interest in deep learning-based prognostic and health management (PHM) for building end-to-end maintenance decision support systems, especially due to the rapid development of autonomous systems. However, the low trustworthiness of PHM hinders its applications in safety-critical assets when handling data from an unknown distribution that differs from the training dataset,… ▽ More There has been a growing interest in deep learning-based prognostic and health management (PHM) for building end-to-end maintenance decision support systems, especially due to the rapid development of autonomous systems. However, the low trustworthiness of PHM hinders its applications in safety-critical assets when handling data from an unknown distribution that differs from the training dataset, referred to as the out-of-distribution (OOD) dataset. To bridge this gap, we propose an uncertainty-informed framework to diagnose faults and meanwhile detect the OOD dataset, enabling the capability of learning unknowns and achieving trustworthy fault diagnosis. Particularly, we develop a probabilistic Bayesian convolutional neural network (CNN) to quantify both epistemic and aleatory uncertainties in fault diagnosis. The fault diagnosis model flags the OOD dataset with large predictive uncertainty for expert intervention and is confident in providing predictions for the data within tolerable uncertainty. This results in trustworthy fault diagnosis and reduces the risk of erroneous decision-making, thus potentially avoiding undesirable consequences. The proposed framework is demonstrated by the fault diagnosis of bearings with three OOD datasets attributed to random number generation, an unknown fault mode, and four common sensor faults, respectively. The results show that the proposed framework is of particular advantage in tackling unknowns and enhancing the trustworthiness of fault diagnosis in safety-critical applications. △ Less

Submitted 8 October, 2021; originally announced November 2021.

Comments: 49 pages

Simulation Based Probabilistic Risk Assessment (SIMPRA): Risk Based Design

Authors: Hamed S Nejad, Tarannom Parhizkar, Ali Mosleh

Abstract: The classical approach to design a system is based on a deterministic perspective where the assumption is that the system and its environment are fully predictable, and their behaviour is completely known to the designer. Although this approach may work fairly well for regular design problems, it is not satisfactory for the design of highly sensitive and complex systems where significant resources… ▽ More The classical approach to design a system is based on a deterministic perspective where the assumption is that the system and its environment are fully predictable, and their behaviour is completely known to the designer. Although this approach may work fairly well for regular design problems, it is not satisfactory for the design of highly sensitive and complex systems where significant resources and even lives are at risk. In addition it can results in extra costs of over-designing for the sake of safety and reliability. In this paper, a risk-based design framework using Simulation Based Probabilistic Risk Assessment (SIMPRA) methodology is proposed. SIMPRA allows the designer to use the knowledge that can be expected to exist at the design stage to identify how deviations can occur; and then apply these high-level scenarios to a rich simulation model of the system to generate detailed scenarios and identify the probability and consequences of these scenarios. SIMPRA has three main modules including Simulator, Planner and Scheduler, and it approach is much more efficient in covering the large space of possible scenarios as compared with, for example, biased Monte Carlo simulations because of the Planner module which uses engineering knowledge to guide the simulation process. The value-added of this approach is that it enables the designer to observe system behaviour under many different conditions. This process will lead to a risk-informed design in which the risk of negative consequences is either eliminated entirely or reduced to an acceptable range. For illustrative purposes, an earth observation satellite system example is introduced. △ Less

Submitted 30 September, 2021; originally announced October 2021.

Convolutional Neural Network Compression through Generalized Kronecker Product Decomposition

Authors: Marawan Gamal Abdel Hameed, Marzieh S. Tahaei, Ali Mosleh, Vahid Partovi Nia

Abstract: Modern Convolutional Neural Network (CNN) architectures, despite their superiority in solving various problems, are generally too large to be deployed on resource constrained edge devices. In this paper, we reduce memory usage and floating-point operations required by convolutional layers in CNNs. We compress these layers by generalizing the Kronecker Product Decomposition to apply to multidimensi… ▽ More Modern Convolutional Neural Network (CNN) architectures, despite their superiority in solving various problems, are generally too large to be deployed on resource constrained edge devices. In this paper, we reduce memory usage and floating-point operations required by convolutional layers in CNNs. We compress these layers by generalizing the Kronecker Product Decomposition to apply to multidimensional tensors, leading to the Generalized Kronecker Product Decomposition (GKPD). Our approach yields a plug-and-play module that can be used as a drop-in replacement for any convolutional layer. Experimental results for image classification on CIFAR-10 and ImageNet datasets using ResNet, MobileNetv2 and SeNet architectures substantiate the effectiveness of our proposed approach. We find that GKPD outperforms state-of-the-art decomposition methods including Tensor-Train and Tensor-Ring as well as other relevant compression methods such as pruning and knowledge distillation. △ Less

Submitted 14 January, 2022; v1 submitted 29 September, 2021; originally announced September 2021.

Human Reliability Analysis for Oil and Gas Operations: Analysis of Existing Methods

Authors: Marilia Ramos, Camille Major, Nsimah Ekanem, Cesar Malpica, Ali Mosleh

Abstract: In the petroleum industry, Quantitative Risk Analysis (QRA) has been one of the main tools for risk management. To date, QRA has mostly focused on technical barriers, despite many accidents having human failure as a primary cause or a contributing factor. Human Reliability Analysis (HRA) allows for the assessment of the human contribution to risk to be assessed both qualitatively and quantitativel… ▽ More In the petroleum industry, Quantitative Risk Analysis (QRA) has been one of the main tools for risk management. To date, QRA has mostly focused on technical barriers, despite many accidents having human failure as a primary cause or a contributing factor. Human Reliability Analysis (HRA) allows for the assessment of the human contribution to risk to be assessed both qualitatively and quantitatively. Most credible and highly advanced HRA methods have largely been developed and applied in support of nuclear power plants control room operations and in context of probabilistic risk analysis. Moreover, many of the HRA methods have issues that have led to inconsistencies, insufficient traceability and reproducibility in both the qualitative and quantitative phases. Given the need to assess human error in the context of the oil industry, it is necessary to evaluate available HRA methodologies and assess its applicability to petroleum operations. Furthermore, it is fundamental to assess these methods against good practices of HRA and the requirements for advanced HRA methods. The present paper accomplishes this by analyzing seven HRA methods. The evaluation of the methods was performed in three stages. The first stage consisted of an evaluation of the degree of adaptability of the method for the Oil and Gas industry. In the second stage the methods were evaluated against desirable items in an HRA method. The higher-ranked methods were evaluated, in the third stage, against requirements for advanced HRA methods. In addition to the methods' evaluation, this paper presents an overview of state-of-the-art discussions on HRA, led by the Nuclear industry community. It remarks that these discussions must be seriously considered in defining a technical roadmap to a credible HRA method for the Oil and Gas industry. △ Less

Submitted 28 September, 2021; originally announced September 2021.

Degradation and Failure Mechanisms of Complex Systems: Principles

Authors: Tarannom Parhizkar, Theresa Stewart, Lixian Huang, Ali Mosleh

Abstract: A cyber physical human complex system failure prevents the accomplishment of the systems intended function. The failure of a complex system could be a breakdown of any system hardware, human related factors, application software, or the interaction between these components. Having knowledge about all these three components would allow us to better understand the behavior, interactions, and the ass… ▽ More A cyber physical human complex system failure prevents the accomplishment of the systems intended function. The failure of a complex system could be a breakdown of any system hardware, human related factors, application software, or the interaction between these components. Having knowledge about all these three components would allow us to better understand the behavior, interactions, and the associated failure mechanisms of the cyber physical human systems as a whole. In this study, degradation mechanisms in these three components are classified and discussed. The main categories are hardware related degradation mechanisms including mechanical, thermal, chemical, electronic and radiation effects degradation mechanisms. In addition to hardware related degradation mechanisms, human failure modes, software errors, and the failures due to cyber physical human interactions are presented and discussed. This paper covers the main types of failure mechanisms in complex systems and is beneficial for develo** conceptual risk and reliability models for complex systems. △ Less

Submitted 5 October, 2021; v1 submitted 23 September, 2021; originally announced September 2021.

Comments: Journal paper

Physics-Informed Deep Learning: A Promising Technique for System Reliability Assessment

Authors: Taotao Zhou, Enrique Lopez Droguett, Ali Mosleh

Abstract: Considerable research has been devoted to deep learning-based predictive models for system prognostics and health management in the reliability and safety community. However, there is limited study on the utilization of deep learning for system reliability assessment. This paper aims to bridge this gap and explore this new interface between deep learning and system reliability assessment by exploi… ▽ More Considerable research has been devoted to deep learning-based predictive models for system prognostics and health management in the reliability and safety community. However, there is limited study on the utilization of deep learning for system reliability assessment. This paper aims to bridge this gap and explore this new interface between deep learning and system reliability assessment by exploiting the recent advances of physics-informed deep learning. Particularly, we present an approach to frame system reliability assessment in the context of physics-informed deep learning and discuss the potential value of physics-informed generative adversarial networks for the uncertainty quantification and measurement data incorporation in system reliability assessment. The proposed approach is demonstrated by three numerical examples involving a dual-processor computing system. The results indicate the potential value of physics-informed deep learning to alleviate computational challenges and combine measurement data and mathematical models for system reliability assessment. △ Less

Submitted 5 September, 2021; v1 submitted 24 August, 2021; originally announced August 2021.

Comments: 29 pages, 15 figures

UHV-CVD Growth of High Quality GeSn Using SnCl4: From Growth Optimization to Prototype Devices

Authors: P. C. Grant, W. Dou, B. Alharthi, J. M. Grant, H. Tran, G. Abernathy, A. Mosleh, W. Du, 5 B. Li, M. Mortazavi, H. A. Naseem, S. Q. Yu

Abstract: The persistent interest of the epitaxy of group IV alloy GeSn is mainly driven by the demand of efficient light source that could be monolithically integrated on Si for mid-infrared Si photonics. For chemical vapor deposition of GeSn, the exploration of parameter window is difficult from the beginning due to its non-equilibrium growth condition. In this work, we demonstrated the effective pathway… ▽ More The persistent interest of the epitaxy of group IV alloy GeSn is mainly driven by the demand of efficient light source that could be monolithically integrated on Si for mid-infrared Si photonics. For chemical vapor deposition of GeSn, the exploration of parameter window is difficult from the beginning due to its non-equilibrium growth condition. In this work, we demonstrated the effective pathway to achieve the high quality GeSn with high Sn incorporation. The GeSn films were grown on Ge-buffered Si via ultra-high vacuum chemical vapor deposition using GeH4 and SnCl4 as precursor gasses. The influence of both SnCl4 flow fraction and growth temperature on the Sn incorporation and material quality were investigated. The key to achieve effective Sn incorporation and high material quality is to explore the proper parameter match between SnCl4 supply and growth temperature, which is also called optimized growth regime. The Sn precipitation is significantly suppressed in optimized growth regime, leading to more Sn incorporation into Ge and enhanced material quality. The prototype GeSn photoconductors were fabricated with typical samples, showing the promising devices applications towards mid-infrared optoelectronics. △ Less

Submitted 5 October, 2018; originally announced October 2018.

Growth of Form in Thin Elastic Structures

Authors: Salem Al Mosleh, Ajay Gopinathan, Christian Santangelo

Abstract: Heterogeneous growth plays an important role in the shape and pattern formation of thin elastic structures ranging from the petals of blooming lilies to the cell walls of growing bacteria. Here we address the stability and regulation of such growth, which we modeled as a quasi-static time evolution of a metric, with fast elastic relaxation of the shape. We consider regulation via coupling of the g… ▽ More Heterogeneous growth plays an important role in the shape and pattern formation of thin elastic structures ranging from the petals of blooming lilies to the cell walls of growing bacteria. Here we address the stability and regulation of such growth, which we modeled as a quasi-static time evolution of a metric, with fast elastic relaxation of the shape. We consider regulation via coupling of the growth law, defined by the time derivative of the target metric, to purely local properties of the shape, such as the local curvature and stress. For cylindrical shells, motivated by rod-like E. \textit{coli}, we show that coupling to curvature alone is generically linearly unstable and that additionally coupling to stress can lead to stably elongating structures. Our approach can readily be extended to gain insights into the general classes of stable growth laws for different target geometries. △ Less

Submitted 1 June, 2018; originally announced June 2018.

Si-based GeSn lasers with wavelength coverage of 2 to 3 μm and operating temperatures up to 180 K

Authors: Joe Margetis, Sattar Al-Kabi, Wei Du, Wei Dou, Yiyin Zhou, Thach Pham, Perry Grant, Seyed Ghetmiri, Aboozar Mosleh, Baohua Li, Jifeng Liu, Greg Sun, Richard Soref, John Tolle, Mansour Mortazavi, Shui-Qing Yu

Abstract: A Si-based monolithic laser is highly desirable for full integration of Si-photonics. Lasing from direct bandgap group-IV GeSn alloy has opened a completely new venue from the traditional III-V integration approach. We demonstrated optically pumped GeSn lasers on Si with broad wavelength coverage from 2 to 3 μm. The GeSn alloys were grown using newly developed approaches with an industry standard… ▽ More A Si-based monolithic laser is highly desirable for full integration of Si-photonics. Lasing from direct bandgap group-IV GeSn alloy has opened a completely new venue from the traditional III-V integration approach. We demonstrated optically pumped GeSn lasers on Si with broad wavelength coverage from 2 to 3 μm. The GeSn alloys were grown using newly developed approaches with an industry standard chemical vapor deposition reactor and low-cost commercially available precursors. The achieved maximum Sn composition of 17.5% exceeded the generally acknowledged Sn incorporation limits for using similar deposition chemistries. The highest lasing temperature was measured as 180 K with the active layer thickness as thin as 260 nm. The unprecedented lasing performance is mainly due to the unique growth approaches, which offer high-quality epitaxial materials. The results reported in this work show a major advance towards Si-based mid-infrared laser sources for integrated photonics. △ Less

Submitted 19 August, 2017; originally announced August 2017.

Comments: 34 pages, 12 figures

MSC Class: 00A79

Nonlinear mechanics of rigidifying curves

Authors: Salem Al Mosleh, Christian Santangelo

Abstract: Thin shells are characterized by a high cost of stretching compared to bending. As a result isometries of the midsurface of a shell play a crucial role in their mechanics. In turn, curves with zero normal curvature play a critical role in determining the number and behavior of isometries. In this paper, we show how the presence of these curves results in a decrease in the number of linear isometri… ▽ More Thin shells are characterized by a high cost of stretching compared to bending. As a result isometries of the midsurface of a shell play a crucial role in their mechanics. In turn, curves with zero normal curvature play a critical role in determining the number and behavior of isometries. In this paper, we show how the presence of these curves results in a decrease in the number of linear isometries. Paradoxically, shells are also known to continuously fold more easily across these rigidifying curves than other curves on the surface. We show how including nonlinearities in the strain can explain this phenomena and demonstrate folding isometries with explicit solutions to the nonlinear isometry equations. In addition to explicit solutions, exact geometric arguments are given to validate and guide our analysis in a coordinate free way. △ Less

Submitted 1 June, 2018; v1 submitted 16 February, 2017; originally announced February 2017.

Journal ref: Phys. Rev. E 96, 013003 (2017)