Gravitational-wave astronomy is an emerging field in observational astrophysics concerned with the study of gravitational signals proposed to exist nearly a century ago by Albert Einstein but only recently confirmed to exist. Such signals were theorized to result from astronomical events such as the collisions of black holes, but they were long thought to be too faint to measure on Earth. In recent years, the construction of extremely sensitive detectors—including the Laser Interferometer Gravitational-Wave Observatory (LIGO) project—has enabled the first direct detections of these gravitational waves, corroborating the theory of general relativity and heralding a new era of astrophysics research. As a result of their extraordinary sensitivity, the instruments used to study gravitational waves are also subject to noise that can significantly limit their ability to detect the signals of interest with sufficient confidence. The detectors continuously record more than 200,000 time series of auxiliary data describing the state of a vast array of internal components and sensors, the environmental state in and around the detector, and so on. This data offers significant value for understanding the nearly innumerable potential sources of noise and ultimately reducing or eliminating them, but it is clearly impossible to monitor, let alone understand, so much information manually. The field of machine learning offers a variety of techniques well-suited to problems of this nature. In this thesis, we develop and present several machine learning–based approaches to automate the process of extracting insights from the vast, complex collection of data recorded by LIGO detectors. We introduce a novel problem formulation for transient noise detection and show for the first time how an efficient and interpretable machine learning method can accurately identify detector noise using all of these auxiliary data channels but without observing the noise itself. We present further work employing more sophisticated neural network–based models, demonstrating how they can reduce error rates by over 60% while also providing LIGO scientists with interpretable insights into the detector’s behavior. We also illustrate the methods’ utility by demonstrating their application to a specific, recurring type of transient noise; we show how we can achieve a classification accuracy of over 97% while also independently corroborating the results of previous manual investigations into the origins of this type of noise. The methods and results presented in the following chapters are applicable not only to the specific gravitational-wave data considered but also to a broader family of machine learning problems involving prediction from similarly complex, high-dimensional data containing only a few relevant components in a sea of irrelevant information. We hope this work proves useful to astrophysicists and other machine learning practitioners seeking to better understand gravitational waves, extremely complex and precise engineered systems, or any of the innumerable extraordinary phenomena of our civilization and universe.

In vivo proton magnetic resonance spectroscopy (1H MRS) is the only method available to measure small-molecule metabolites in living human tissue, including the brain, without ionizing radiation or invasive medical procedures. Despite its attendant potential for supporting clinical diagnostics in a range of neurological and psychiatric conditions, the metabolite concentration estimates produced by 1H-MRS experiments, and therefore their sensitivity and specificity to any particular biological phenomenon under study, are readily distorted by a number of confounds. These include but are not limited to static and radiofrequency field characteristics, signal relaxation dynamics, macromolecule and lipid contributions to the spectral baseline, spectral fitting artifacts, and other uncontrolled idiosyncrasies of 1H-MRS data acquisition, processing, and quantification. Using 1H-MRS data obtained via 3-Tesla and 7-Tesla magnetic resonance (MR) scanners from healthy controls, individuals with progressive and relapsing-remitting multiple sclerosis (MS), and individuals with post-traumatic stress disorder (PTSD) and/or major depressive disorder (MDD), this work therefore aims to build and apply a framework for quantifying and thereby reducing such confounds introduced to 1H-MRS estimates of in vivo metabolite concentrations at the steps of data processing and quantification, with an ultimate aim to maximizing the potential of 1H MRS for supporting sensitive and specific clinical diagnosis of neurological or psychiatric disease. The steps examined include spectral quantification by linear combination modeling (Chapter 2), absolute quantification by internal concentration referencing (Chapter 3), and cross-sectional statistical analysis of results (Chapters 4 and 5). Chapter 2 designs and implements a graphical user interface (GUI)-supported validation pipeline for measuring how data quality, spectral baseline, and baseline model affect the precision and accuracy of 1H-MR spectral quantification by linear combination modeling. This validation pipeline is then used to show that spectral data quality indices signal to noise ratio (SNR) and full width at half maximum (FWHM) interact with spectral baseline to influence not only the precision but also the accuracy of resultant metabolite concentration estimates, with fit residuals poorly indicative of true fit error and spectral baselines modeled as regularized cubic splines not significantly outperformed by those employing simulated macromolecules. A novel method for extending the commonly used spectral quantification precision estimate Cramér-Rao Lower Bound (CRLB) to incorporate considerations of continuous and piecewise polynomial baseline shapes is therefore presented, tested, and similarly integrated into a GUI-supported toolkit to improve the correspondence between estimated CRLB and metabolite fit error variability when this now empirically justified approach to spectral baseline modeling is used. In Chapter 3, age- and disease-associated differences in transverse (T2) water signal relaxation measured at 7 Tesla in the prefrontal cortex of individuals with progressive (N=21) relative to relapsing-remitting (N=26) or no (N=25) multiple sclerosis are shown to influence absolute quantification of metabolite concentrations by internal referencing to water. In Chapter 4, these findings from Chapters 2 and 3 are used to justify an evidence-based 1H-MR spectral processing and quantification protocol that focuses optimization efforts on baseline modeling approach and references metabolite concentration estimates to internal creatine instead of water. When this protocol is applied to 7-Tesla prefrontal cortex 1H-MR spectra from the aforementioned multiple sclerosis and control cohorts, it supports metabolite concentration estimates that, in the absence of any additional supporting data, inform supervised-learning-enabled identification of progressive multiple sclerosis at nearly 80% held-out validation sensitivity and specificity. Finally, in Chapter 5, the same processing, quantification, and machine-learning pipeline employed in Aim 3 is independently applied to a new set of 7-Tesla prefrontal cortex 1H-MRS raw data from an entirely different cohort of individuals with (N=20) and without (N=18) PTSD and/or comorbid or primary MDD. Here the processing, quantification, and statistics procedures designed using lessons in Chapters 2 and 3 and optimized for classifying multiple sclerosis phenotype in Chapter 4 generalize directly to metabolite-only classification of PTSD and/or MDD with sensitivity and specificity similarly near to or greater than 80%. In both Chapters 4 and 5, supervised learning avoids dimensionally reducing metabolite feature sets in order to pinpoint the specific metabolites most informative for identifying each disease group. Taken together, these findings justify the potential and continued development of 1H MRS, at least as applied in the human brain and especially as supported by multivariate approaches including supervised learning, as an auxiliary or mainstay of clinical diagnostics for neurological or psychiatric disease.

In the early stages of new product development, project selection is dominantly based on managerial intuition, rather than on analytic approaches. As much as 90% of all product ideas are rejected before they are formally assessed. However, to date, little is known about the product screening heuristics and screening criteria managers use: it has been suggested that their decision process resembles the "fast and frugal" heuristics identified in recent psychological research, but no empirical research exists. A major part of the product innovation pipeline is thus poorly understood. This research contributes to closing this gap. It uses cognitive task analysis for an in-depth analysis of the new product screening heuristics of twelve experienced decision makers in 66 decision cases. Based on the emerging data, an integrated model of their project screening heuristics is created. Results show that experts adapt their heuristics to the decision at hand. In doing so, they use a much smaller set of decision criteria than discussed in the product development literature. They also combine heuristics into decision approaches that are simple, but more complex than "fast and frugal" strategies. By opening the black box of project screening this research enables improved project selection practices.

This work presents a new digital cooking process that utilizes heat from lasers to cookfood products. Unlike conventional cooking appliances, which heat an entire area by some uniform amount, lasers are unique in that they provide precision directional heating, they have a small form factor, and they are highly controllable using software. Lasers—as a cooking appliance—are of particular interest since they have a heating resolution that is on the same order of magnitude as the deposition path of a 3D food printer. While food printing is great for customized meal creation, we can print foods to millimeter resolution but we lack the ability to cook at this same resolution. Here, I primarily focus on the characterization of three different types of lasers: (1) a blue laser operating at 445 nm, (2) a near-infrared (NIR) laser operating at 980 nm, and (3) a mid-infrared (MIR) laser operating at 10.6 µm. Initial cooking apparatuses used a set of mirror galvanometers to direct visible blue light for cooking, then future iterations relied on the movement of a 3-axis gantry. Both blue and NIR lasers are diode lasers that can be mounted on a machine and the MIR laser is a standalone CO₂ gas laser. I characterize the heating behavior of the aforementioned lasers using dough, salmon, and chicken as model food systems. Different modes of cooking can be achieved by changing the wavelength of the light: infrared (IR) lasers are more well-suited for non-enzymatic browning and blue lasers are best for subsurface cooking (i.e. starch gelatinization of dough, protein denaturation of salmon and chicken). Precision “pulsed heating” with lasers also allows one to achieve food safe temperatures with greater accuracy and reduces overcooking, which leads to more moist food samples. Laser-baked dough can also achieve starch gelatinization. Food safe temperatures and browning can be achieved in dough, salmon, and chicken. Lastly, color development—as a result of laser exposure—is similar to conventionally cooked foods.

No description available.

In this dissertation I study the benefits that machine learning can bring to problems of Sustainable Development in the field of hydrology. Specifically, in Chapter 1 I investigate how predictable groundwater depletion is across India and to what extent we can learn from the model’s predictions about underlying drivers. In Chapter 2, I joined a competition to predict the amount of water in snow in the western United States using satellite imagery and convolutional neural networks. Lastly, in Chapter 3 I examine how cloud cover impacts the machine learning model’s predictions and explore how cloudiness impacts the successes and limitation of the popular uncertainty quantification method known as Monte Carlo dropout. Food production in many parts of the world relies on groundwater resources. In many regions, groundwater levels are declining due to a combination of anthropogenic abstraction, localized meteorological and geological characteristics, and climate change. Groundwater in India is characteristic of this global trend, with an agricultural sector that is highly dependent on groundwater and increasingly threatened by abstraction far in excess of recharge. The complexity of inputs makes groundwater depletion highly heterogeneous across space and time. However, modeling this heterogeneity has thus far proven difficult. In Chapter 1 using random forest models and high-resolution feature importance methods, we demonstrate a recent shift in the predictors of groundwater depletion in India and show an improved ability to make predictions at the district-level across seasons. We find that, as groundwater depletion begins to accelerate across India, deep-well irrigation use becomes 250% more important from 1996-2014, becoming the most important predictor of depletion in the majority of districts in northern and central India. At the same time, even many of the districts that show gains in groundwater levels show an increasing importance of deep irrigation. Analysis shows widespread decreases in crop yields per unit of irrigation over our time period, suggesting decreasing marginal returns for the largely increasing quantities of groundwater irrigation used. Because anthropogenic and natural drivers of groundwater recharge are highly localized, understanding the relationship between multiple variables across space and time is inferentially challenging, yet extremely important. Our granular, district-focused models of groundwater depletion rates can inform decision-making across diverse hydrological conditions and water use needs across space, time, and groups of constituents. In Chapter 2 I reflect on competing in the U.S. Bureau of Reclamation’s snow water equivalent prediction competition (Snowcast Showdown). This project was a joint effort with Isabella Smythe and we ended the competition scoring roughly 45th out of over 1000 teams on the public leaderboard. In this chapter I outline our approach and discuss the competition format, model building, and examine alternative approaches taken by other competitors. Similarly I consider the success and limitations of our own satellite-based approach and consider future improvements to iterate upon our model. In Chapter 3 I study the black-box deep learning model built on MODIS imagery to estimate snow water equivalent (SWE) made for the competition discussed in Chapter 2. Specifically, I here investigate a major component of uncertainty in my remotely-sensed images: cloud cover which completely disrupts viewing of the surface in the visible spectrum. To understand the impact of cloud-driven missingness, I document how and where clouds occur in the dataset. I then use Monte Carlo dropout - a popular method of quantifying uncertainty in deep learning models - to learn how well the method captures the aleatoric errors unique to remote sensing with cloud cover. Next, I investigate how the underlying filters of the convolutional neural network appear using the guided backprop technique and draw conclusions regarding what features in the images the model was using to make its predictions. Lastly, I investigate what forms of validation best estimated the true generalization error in Chapter 2 using ordinary least squares (OLS) and the elastic-net technique. These three chapters show that machine learning has an important place in the future of hydrology, however the tools that it brings are still difficult to interpret. Moreover, future work is still needed to bring these predictive advancements to scientific standards of understanding. This said, the increases to accuracy brought by the new techniques can currently make a difference to people’s lives who will face greater water scarcity as climate change accelerates.

In recent years, due to the increasing efficiency and availability of information technologies for collecting massive amounts of data (e.g., smart meters and sensors), a variety of advanced technologies and decision-making strategies in the civil engineering sector have shifted in leaps and bounds to a data-driven manner. While there is still no consensus in industry and academia on the latest advances, challenges, and trends in some innovative data-driven methods related to, e.g., deep learning and neural networks, it is undeniable that these techniques have been proven to be considerably effective in helping our academics and engineers solve many real-life tasks related to the smart city framework. This dissertation systematically presents the investigation and development of the cutting-edge data-driven methods related to two specific areas of civil engineering, namely, Residential Electricity Modeling (REM) and Structural Health Monitoring (SHM). For both components, the presentation of this dissertation starts with a brief review of classical data-driven methods used in particular problems, gradually progresses to an exploration of the related state-of-the-art technologies, and eventually lands on our proposed novel data-driven strategies and algorithms. In addition to the classical and state-of-the-art modeling techniques focused on these two areas, this dissertation also put great emphasis on the proposed effective feature extraction and selection approaches. These approaches are aimed to optimize model performance and to save computational resources, for achieving the ideal characterization of the information embedded in the collected raw data that is most relevant to the problem objectives, especially for the case of modeling deep neural networks. For the problems on REM, the proposed methods are validated with real recorded data from multi-family residential buildings, while for SHM, the algorithms are validated with data from numerically simulated systems as well as real bridge structures.

Cutting tools in a manufacturing system play a significant role in metal cutting process. Cutting tools arc consumable. Tool wear affects the outcomes of machining processes such as machined· surface quality and it is normally used as an indicating parameter for tool life. It is crucial to calculate and record the wear size of different cutting tools. The tool wear is classified into many different categories according to its existing location and wear patterns. Flank wear and crater wear are the two most common types of tool wear which are used to assess the tool's life. The measurements of flank and crater wear in cutting tools have been extensively studied. There are still many challenges when these research results are applied practically. Manufacturing industry demands· accurate and rapid methods for the tool wear measurement. There are two primary objectives in this research. The first is to develop a new tool wear measurement technology by using the active contour model based image processing method for the flank wear measurement. A MATLAB program is developed to verify the suggested image processing algorithm. Many cutting experiments were conducted with different tools on a CNC machine tool. The experimental results show that the developed technology is feasible and can be used to measure the tool wear area. It needs to be noted that this method. can only extract the wear area. This method is not able to estimate the depth of the crater wear. The second objective is to develop a method for crater wear measurement by use of developed stereo vision system. This system consists of a single camera with necessary lighting devices and fixtures. A MATLAB based software is developed to estimate and represent the volume of tool wear. The proposed algorithm and feasibility of the system for the crater wear (3D tool wear) is discussed. Its effectiveness is verified in this research.

There is a growing number of applications of machine learning and deep learning in quantitative and computational finance. In this thesis, we focus on two of them. In the first application, we employ machine learning and deep learning in derivative pricing. The models considering jumps or stochastic volatility are more complicated than the Black-Merton-Scholes model and the derivatives under these models are harder to be priced. The traditional pricing methods are computationally intensive, so machine learning and deep learning are employed for fast pricing. I n Chapter 2, we propose a method for pricing American options under the variance gamma model. We develop a new fast and accurate approximation method inspired by the quadratic approximation to get rid of the time steps required in finite difference and simulation methods, while reducing the error by making use of a machine learning technique on pre-calculated quantities. We compare the performance of our method with those of the existing methods and show that this method is efficient and accurate for practical use. In Chapters 3 and 4, we propose unsupervised deep learning methods for option pricing under Lévy process and stochastic volatility respectively, with a special focus on barrier options in Chapter 4. The unsupervised deep learning approach employs a neural network as the candidate option surface and trains the neural network to satisfy certain equations. By matching the equation and the boundary conditions, the neural network would yield an accurate solution. Special structures called singular terms are added to the neural networks to deal with the non-smooth and discontinuous payoff at the strike and barrier levels so that the neural networks can replicate the asymptotic behaviors of options at short maturities. Unlike supervised learning, this approach does not require any labels. Once trained, the neural network solution yields fast and accurate option values. The second application focuses on financial time series simulation utilizing deep learning techniques. Simulation extends the limited real data for training and evaluation of trading strategies. It is challenging because of the complex statistical properties of the real financial data. In Chapter 5, we introduce two generative adversarial networks, which utilize the convolutional networks with attention and the transformers, for financial time series simulation. The networks learn the statistical properties in a data-driven manner and the attention mechanism helps to replicate the long-range dependencies. The proposed models are tested on the S&P 500 index and its option data, examined by scores based on the stylized facts and are compared with the pure convolutional network, i.e. QuantGAN. The attention-based networks not only reproduce the stylized facts, including heavy tails, autocorrelation and cross-correlation, but also smooth the autocorrelation of returns.

Autonomous Driving Systems (ADSs) have seen rapid progress in recent years. To ensure the safety and reliability of these systems, extensive testing is being conducted before their future mass deployment. One approach is to test ADSs directly on the road, but it is incredibly costly to cover all rare corner cases. Thus, a popular complementary approach is to evaluate an ADS’s performance in a simulator. Such method is called simulation based testing. However, randomly testing ADSs in simulation is still not efficient enough and the testing results might not transfer to the real-world. This dissertation underscores that the cornerstone of efficient simulation testing lies in crafting optimal testing scenarios. We delineate several pivotal properties for these scenarios: they should induce ADS misbehavior, exhibit diversity, manifest realism, and adhere to user specified rules (e.g., following traffic rules). Subsequent to this identification, our research delves into methodologies to enhance one or more of these properties of the generated scenarios. Specifically, we embark on two distinct lines of approach. First, we develop advanced search strategies to unearth diverse scenarios that provoke ADS to misbehave. Second, we harness the potential of deep generative models to produce scenarios that are both realistic and in compliance with user specified rules. Because of the need for efficiently testing end-to-end behaviors of ADSs against complex, real-world corner cases, we propose AutoFuzz, a novel fuzz testing technique, which can leverage widely-used driving simulators’ API grammars to generate complex driving scenarios. In order to find misbehavior-inducing scenarios, which are very rare, we propose a learning based search method to optimize AutoFuzz. In particular, our method trains a neural network to select and mutate scenarios sampled from an evolutionary search method. AutoFuzz shows promises in efficiently identifying traffic violations for the given ADSs under test. Although AutoFuzz is good at finding violations, as a black-box method, it is agnostic of the cause of the violations. In the second project, we focus on finding violations caused by the failure of fusion component, which fuses the inputs of multiple sensors and provides the ADS a more reliable understanding of the surroundings. In particular, we identify that the fusion component of an industry-grade ADAS can fail to trust the more reliable input sensor and thus lead to a collision. We define misbehavior caused by such a failure as "fusion error". In order to efficiently find fusion errors, we propose a fuzzing framework, named FusED, that uses a novel evolutionary-based search method with objective promoting fusion output to deviate from sensor input. We show that FusED can efficiently reveal fusion errors for an industry-grade ADAS. One issue with the generated scenarios by AutoFuzz or FusED (or any other search based methods) is that all the NPC vehicles are controlled by some low-level controllers, whose behaviors are different from human drivers. This poses a difficulty in transferring the found violations into real world. Some recent work tries to address this problem by using deep generative models. However, the scenarios cannot be easily controlled which is not desirable for users to customize the testing scenarios. As both realism and controllability of the generated traffic are desirable, we propose a novel method called Controllable Traffic Generation (CTG) that achieves both properties simultaneously. In order to preserve realism, we propose a conditional, dynamic enforced diffusion model. In order to satisfy controllability, we propose using a kind of "traffic language" called Signal Temporal Logic (STL) to specify what we want in traffic scenarios (e.g., following road rules). We then leverage STL to guide the conditional diffusion model for generating realistic and controllable traffic. Although CTG can generate realistic and controllable traffic, it still requires domain expertise to specify the STL based loss function. Besides, it models traffic participants independently, resulting in sub-optimal agents interaction modeling. In order to address these issues, we developed CTG++ which enables a user to use language to generate realistic traffic scenario. In particular, we proposed to use GPT4 to translate a command in natural language into a loss function in code. We then use the loss function to guide a scene-level diffusion model, which considers all the vehicles jointly, to generate traffic satisfying the command. We have found that CTG++ can generate query (in natural language)-compliant and realistic traffic simulation. In summary, our four projects discussed in this thesis have solved important problems in efficiently testing ADSs and have had significant influence in the advancement of ADS. Besides, the models and empirical studies we performed can be applicable to other testing and behavior generation problems, such as general ML-based software testing, and multi-agent behavior planning and prediction. I hope this thesis can serve as an inspiration to anyone who is interested in the exciting field of ADS testing and development, and contribute to the realization of the full automation of driving.