Browsing by Subject "Reliability"

I develop an empirical model that estimates a firm-specific accrual noise ratio (ANR), an operational and statistically grounded measure of accrual reliability, and test the measure's construct validity. The model allows accrual reliability to vary across firms, which is particularly important because many reliability determinants vary in cross-section. Unlike metrics that measure relative perceived reliability, ANR measures accrual reliability independent of the perceptions of investors, creditors or auditors. I find that ANR relates in expected ways with multiple proxies of accounting reliability, that ANR's relation with the proxies of other accounting constructs is consistent with theory, and that ANR's sensitivity to percentage changes of accrual components is consistent with a subjective ordinal ranking of the components' reliability from prior literature.

In this dissertation, we develop a novel methodology for characterizing and simulating nonstationary, full-field, stochastic turbulent wind fields.

In this new method, nonstationarity is characterized and modeled via temporal coherence, which is quantified in the discrete frequency domain by probability distributions of the differences in phase between adjacent Fourier components.

The empirical distributions of the phase differences can also be extracted from measured data, and the resulting temporal coherence parameters can quantify the occurrence of nonstationarity in empirical wind data.

This dissertation (1) implements temporal coherence in a desktop turbulence simulator, (2) calibrates empirical temporal coherence models for four wind datasets, and (3) quantifies the increase in lifetime wind turbine loads caused by temporal coherence.

The four wind datasets were intentionally chosen from locations around the world so that they had significantly different ambient atmospheric conditions.

The prevalence of temporal coherence and its relationship to other standard wind parameters was modeled through empirical joint distributions (EJDs), which involved fitting marginal distributions and calculating correlations.

EJDs have the added benefit of being able to generate samples of wind parameters that reflect the characteristics of a particular site.

Lastly, to characterize the effect of temporal coherence on design loads, we created four models in the open-source wind turbine simulator FAST based on the \windpact turbines, fit response surfaces to them, and used the response surfaces to calculate lifetime turbine responses to wind fields simulated with and without temporal coherence.

The training data for the response surfaces was generated from exhaustive FAST simulations that were run on the high-performance computing (HPC) facilities at the National Renewable Energy Laboratory.

This process was repeated for wind field parameters drawn from the empirical distributions and for wind samples drawn using the recommended procedure in the wind turbine design standard \iec.

The effect of temporal coherence was calculated as a percent increase in the lifetime load over the base value with no temporal coherence.

Deep Neural Networks (DNNs) have driven the performance of computer vision to new heights, which has led to them to being rapidly integrated into many of our real-world systems. Meanwhile, the majority of research on DNNs remains focused on enhancing accuracy and efficiency. Furthermore, the evaluation protocols used to quantify performance generally assume idealistic operating conditions that do not well-emulate realistic environments. For example, modern benchmarks typically have balanced class distributions, ample training data, consistent object scale, minimal noise, and only test on inputs that lie within the training distribution. As a result, we are currently integrating these naive and under-tested models into our trusted systems! In this work, we focus on the robustness of DNN-based vision models, seeking to understand their vulnerabilities to non-ideal deployment data. The rallying cry of our research is that before these models are deployed into our safety-critical applications (e.g., autonomous vehicles, defense technologies), we must attempt to anticipate, understand, and address all possible vulnerabilities. We begin by investigating a class of malignant inputs that are specifically designed to fool DNN models. We conduct this investigation by taking on the perspective of an adversary who wishes to attack a pretrained DNN by adding (nearly) imperceptible noise to a benign input to fool a downstream model. While most adversarial literature focuses on image classifiers, we seek to understand the feasibility of attacks on other tasks such as video recognition models and deep reinforcement learning agents. Sticking to the theme of \textit{realistic} vulnerabilities, we primarily focus on black-box attacks in which the adversary does not assume knowledge of the target model's architecture and parameters. Our novel attack algorithms achieve surprisingly strong effectiveness, thus uncovering new serious potential security risks.

While malignant adversarial inputs represent a critical vulnerability, they are still a fairly niche issue in the context of all problematic inputs for a DNN. In the second phase of our work, we turn our attention to the open-set vulnerability. Here, we acknowledge that during deployment, models may encounter novel classes from outside of their training distribution. Again, the majority of works in this area only consider image classifiers for their simplicity. This motivates us to study the more complex and practically useful open-set object detection problem. We address this problem in two phases. First, we create a tunable class-agnostic object proposal network that can be easily adapted to suit a variety of open-set applications. Next, we define a new Open-Set Object Detection and Discovery (OSODD) task that emphasizes both known and unknown object detection with class-wise separation. We then devise a novel framework that combines our tunable proposal network with a powerful transformer-based foundational model, which achieves state-of-the-art performance on this challenging task.

We conclude with a feasibility study of inference-time dynamic Convolutional Neural Networks (CNNs). We argue that this may be an exciting potential solution for improving robustness to natural variations such as changing object scale, aspect ratio, and surrounding contextual information. Our preliminary results indicate that different inputs have a strong preference for different convolutional kernel configurations. We show that by allowing just four layers of common off-the-shelf CNN models to have dynamic convolutional stride, dilation, and size, we can achieve remarkably high levels of accuracy on classification tasks.

Inter-Vehicle Communication (IVC) is a vital part of Intelligent Transportation System (ITS), which has been extensively researched in recent years. Dedicated Short Range Communication (DSRC) is being seriously considered by automotive industry and government agencies as a promising wireless technology for enhancing transportation safety and efficiency of road utilization. In the DSRC based vehicular ad hoc networks (VANETs), the transportation safety is one of the most crucial features that needs to be addressed. Safety applications usually demand direct vehicle-to-vehicle ad hoc communication due to a highly dynamic network topology and strict delay requirements. Such direct safety communication will involve a broadcast service because safety information can be beneficial to all vehicles around a sender. Broadcasting safety messages is one of the fundamental services in DSRC. In order to provide satisfactory quality of services (QoS) for various safety applications, safety messages need to be delivered both timely and reliably. To support the stringent delay and reliability requirements of broadcasting safety messages, researchers have been seeking to test proposed DSRC protocols and suggesting improvements. A major hurdle in the development of VANET for safety-critical services is the lack of methods that enable one to determine the effectiveness of VANET design mechanism for predictable QoS and allow one to evaluate the tradeoff between network parameters. Computer simulations are extensively used for this purpose. A few analytic models and experiments have been developed to study the performance and reliability of IEEE 802.11p for safety-related applications. In this thesis, we propose to develop detailed analytic models to capture various safety message dissemination features such as channel contention, backoff behavior, concurrent transmissions, hidden terminal problems, channel fading with path loss, multi-channel operations, multi-hop dissemination in 1-Dimentional or 2-Dimentional traffic scenarios. MAC-level and application-level performance metrics are derived to evaluate the performance and reliability of message broadcasting, which provide insights on network parameter settings. Extensive simulations in either Matlab or NS2 are conducted to validate the accuracy of our proposed models.