93887 research outputs found
Sort by
Retrospective Approximation for Smooth Stochastic Optimization
No full textStochastic Gradient Descent (SGD) is a widely-used iterative algorithm for solving stochastic optimization problems for a smooth (and possibly non-convex) objective function via queries from a first-order stochastic oracle. In this dissertation, we critically examine SGD’s choice of executing a single step as opposed to multiple steps between subsample updates. Our investigation leads naturally to generalizing SG into Retrospective Approximation (RA) where, during each iteration, a deterministic solver executes possibly multiple steps on a subsampled deterministic problem and stops when further solving is deemed unnecessary from the standpoint of statistical efficiency. RA thus leverages what is appealing for implementation – during each iteration, a solver, e.g., L-BFGS with backtracking line search is used, as is, and the subsampled objected function is solved only to the extent necessary. We develop a complete theory using relative errorof the observed gradients as the principal object, demonstrating that almost sure and L1 consistency of RA are preserved under especially weak conditions when sample sizes are increased at appropriate rates. We also characterize the iteration and oracle complexity (for linear and sub-linear solvers) of RA, and identify two practical termination criteria, one of which we show leads to optimal complexity rates. The message from extensive numerical experiments is that the ability of RA to incorporate existing second-order deterministic solvers in a strategic manner is useful both in terms of algorithmic trajectory as well as from the standpoint of dispensing with hyper-parameter tuning
Hardware/Software Co-Design for Keyword Spotting on Edge Devices
No full textThe introduction of artificial neural networks (ANNs) to speech recognition applications has sparked the rapid development and popularization of digital assistants. These digital assistants perform keyword spotting (KWS), constantly monitoring the audio captured by a microphone for a small set of words or phrases known as keywords. Upon recognizing a keyword, a larger audio recording is saved and processed by a separate, more complex neural network. More broadly, neural networks in speech recognition have popularized voice as means of interacting with electronic devices, sparking an interest in individuals using speech recognition in their own projects. However, while large companies have the means to develop custom neural network architectures alongside proprietary hardware platforms, such development precludes those lacking similar resources from developing efficient and effective neural networks for embedded systems. While small, low-power embedded systems are widely available in the hobbyist space, a clear process is needed for developing a neural network that accounts for the limitations of these resource-constrained systems. In contrast, a wide variety of neural network architectures exists, but often little thought is given to deploying these architectures on edge devices.This thesis first presents an overview of audio processing techniques, artificial neural network fundamentals, and machine learning tools. A summary of a set of specific neural network architectures is also discussed. Finally, the process of implementing and modifying these existing neural network architectures and training specific models in Python using TensorFlow is demonstrated. The trained models are also subjected to post-training quantization to evaluate the effect on model performance. The models are evaluated using metrics relevant to deployment on resource-constrained systems, such as memory consumption, latency, and model size, in addition to the standard comparisons of accuracy and parameter count. After evaluating the models and architectures, the process of deploying one of the trained and quantized models is explored on an Arduino Nano 33 BLE using TensorFlow Lite for Microcontrollers and on a Digilent Nexys 4 FPGA board using CFU Playground
Optimal Ordering Policies for Supply Networks with Disruptions
No full textAs the economy recovered with the winding down of the pandemic, businesses with complex supply chains could not bring their inventories back to optimal levels as their production was susceptible to disruption due to supply outages. Deriving optimal ordering policies as a way to mitigate the impact of production disruption represents a challenge in multi-stage decision problems given the complexity of the network and the uncertainty in the demand.In the first part of the thesis, we formulate a stochastic inventory control problem for a general supply network model. Using the Bellman’s recursion and properties of the cost function at each stage, we characterize the optimal request decision as a threshold policy where the threshold computation is based on the marginal cost. Lastly, we validate that the policy developed minimizes the inventory cost and meets an exogenous random demand. However, the policy does not guarantee that the inventory level for each firm satisfies the constraints when a supply disruption occurs in the network.In the second part of the thesis, we consider a serial network in which firms engage in production subject to disruption risk and they look to maximize their profit. We propose an algorithm to characterize a stationary optimal policy based on the closed-form solutions obtained from a discounted finite horizon problem for profit maximization. Finally, by computing the policy proposed as a function of the tier’s location and its disruption probability, we provide simulation results of the disruption effect in the supply network
Methods for Ultrasound Imaging of Multi- Layered Objects Based on Collimated Beam Systems
No full textNon-destructive characterization of multi-layered structures that can be accessed from only a single side is important for applications such as well-bore integrity inspection. Collimated beam ultrasound systems are a technology for imaging inside multi-layered structures such as geothermal wells. These systems work by using a collimated narrow-band ultrasound transmitter that can penetrate through multiple layers of heterogeneous material. A series of measurements can then be made at multiple transmit frequencies. However, commonly used reconstruction algorithms such as Synthetic Aperture Focusing Technique (SAFT) tend to produce poor quality reconstructions for these systems both because they do not model collimated beam systems and they do not jointly reconstruct the multiple frequencies.In this thesis, we first propose a multi-frequency ultrasound model-based iterative reconstruction (UMBIR) algorithm designed for multi-frequency collimated beam ultrasound systems. The combined system targets reflective imaging of heterogeneous, multi-layered structures. For each transmitted frequency band, we introduce a physics-based forward model to accurately account for the propagation of the collimated narrow-band ultrasonic beam through the multi-layered media. We then show how the joint multi-frequency UMBIR reconstruction can be computed by modeling the direct arrival signals, detector noise, and incorporating a spatially varying image prior.We also propose a ringing artifact reduction method for ultrasound image reconstruction that uses a multi-agent consensus equilibrium (RARE-MACE) framework. Our approach integrates a physics-based forward model that accounts for the propagation of a collimated ultrasonic beam in multi-layered media, a spatially varying image prior, and a denoiser designed to suppress the ringing artifacts that are characteristic of reconstructions from high-fractional bandwidth ultrasound sensor data.Results using both simulated and experimental data indicate that multi-frequency UMBIR reconstruction yields much higher reconstruction quality than either single frequency UMBIR or SAFT. In addition, our results demonstrate the capability of our RARE-MACE method to suppress ringing artifacts and substantially improve the image quality over single frequency UMBIR and SAFT
A Parallel Computing Approach for Identifying Retinitis Pigmentosa Modifiers in Drosophila Using Eye Size and Gene Expression Data
No full textFor many years, researchers have developed ways to diagnose degenerative disease in the retina by utilizing multiple gene analysis techniques. Retinitis pigmentosa (RP) disease can cause either partially or totally blindness in adults. For that reason, it is crucial to find a way to pinpoint the causes in order to develop a proper medication or treatment. One of the common methods is genome-wide analysis (GWA). However, it cannot fully identify the genes that are indirectly related to the changes in eye size. In this research, RNA sequencing (RNA-seq) analysis is used to link the phenotype to genotype, creating a pool of candidate genes that might associate with the RP. This will support future research in finding a therapy or treatment to cure such disease in human adults.Using the DrosophilaGenetic Reference Panel (DGRP) – a gene reference panel of fruit fly – two types of datasets are involved in this analysis: eye-size data and gene expression data with two replicates for each strain. This allows us to create a phenotype-genotype map. In other words, we are trying to trace the genes (genotype) that exhibit the RP disease guided by comparing their eye size (phenotype). The basic idea of the algorithm is to discover the best replicate combination that maximizes the correlation between gene expression and eye-size. Since there are 2N possible replicate combinations, where N is the number of selected strains, the original implementation of sequential algorithm was computationally intensive.The original idea of finding the best replicate combination was proposed by Nguyen et al. (2022). In this research, however, we restructured the algorithms to distribute the tasks of finding the best replicate combination and run them in parallel. The implementation was done using the R programming language, utilizing doParallel and foreach packages, and able to execute on a multicore machine. The program was tested on both a laptop and a server, and the experimental results showed an outstanding improvement in terms of the execution time. For instance, while using 32 processes, the results reported up to 95% reduction in execution time when compared with the sequential version of the code. Furthermore, with the increment of computational capabilities, we were able to explore and analyze more extreme eye-size lines using three eye-size datasets representing different phenotype models. This further improved the accuracy of the results where the top candidate genes from all cases showed connection to RP
Computational Mechanobiology Model Evaluating Healing of Postoperative Cavities Following Breast-Conserving Surgery
No full textBreast cancer is the most commonly diagnosed cancer type worldwide. Given high survivorship, increased focus has been placed on long-term treatment outcomes and patient quality of life. While breast-conserving surgery (BCS) is the preferred treatment strategy for early-stage breast cancer, anticipated healing and breast deformation (cosmetic) outcomes weigh heavily on surgeon and patient selection between BCS and more aggressive mastectomy procedures. Unfortunately, surgical outcomes following BCS are difficult to predict, owing to the complexity of the tissue repair process and significant patient-to-patient variability. To overcome this challenge, we developed a predictive computational mechanobiological model that simulates breast healing and deformation following BCS. The coupled biochemical-biomechanical model incorporates multi-scale cell and tissue mechanics, including collagen deposition and remodeling, collagen-dependent cell migration and contractility, and tissue plastic deformation. Available human clinical data evaluating cavity contraction and histopathological data from an experimental porcine lumpectomy study were used for model calibration. The computational model was successfully fit to data by optimizing biochemical and mechanobiological parameters through the Gaussian Process. The calibrated model was then applied to define key mechanobiological parameters and relationships influencing healing and breast deformation outcomes. Variability in patient characteristics including cavity-to-breast volume percentage and breast composition were further evaluated to determine effects on cavity contraction and breast cosmetic outcomes, with simulation outcomes aligning well with previously reported human studies. The proposed model has the potential to assist surgeons and their patients in developing and discussing individualized treatment plans that lead to more satisfying post-surgical outcomes and improved quality of life
Optimization of Steelmaking Processes in an Electric ARC Furnace
No full textThe Electric Arc Furnace is being used quite frequently and significantly in today’s environment due to its energy saving and improved product quality features. It is even more important to study more about optimizing the processes involved in the EAF in order to capitalize on its positive features and thus able to contribute to the world in our bid to fight global climate change.During the refining stage of the Electric Arc Furnace (EAF) operation, molten steel is stirred to facilitate steel/slag reactions and the removal of impurities, which determines the quality of the steel. The stirring process is driven by the injection of oxygen, which is carried out by burners operating in lance mode. In this study, a computational platform is used to analyze the flow dynamics produced during the stirring of the steel bath in an industry-scale EAF. Namely, nonreacting, three-dimensional, transient simulations of the liquid bath stirred by oxygen injection are carried out to analyze the mixing process. The CFD domain includes the liquid bath and the oxygen injected by three coherent jets. The study includes a baseline case, where the oxygen is injected at 1000 SCFM in all the burners. Two sets of cases are also included: first set considers cases where oxygen is injected at a reduced and at an increased uniform flow rate: 750 and 1250 SCFM, respectively. Second set considers cases with non-uniform injection rates in each burner, which keep the same total flow rate of the baseline case, 3000 SCFM. The analysis is also quantified by defining two variables: the mixing time and the standard deviation of the flow velocity. Results indicate that the mixing rate of the bath is determined by flow dynamics near the injection cavities, and that the formation of very low velocity regions or ‘dead zones’ at the center of the furnace and at the balcony regions prevent the flow mixing. All the non-uniform injection cases reduce the mixing time obtained in the baseline case. The melting of HBI/scrap in an electric arc furnace (EAF) is studied by using a computational fluid dynamics (CFD) platform. A previously developed scrap melting model is extended to investigate the different physical phenomena involved in melting of HBI/scrap charges. The CFD platform merges three computational models, namely a scrap/HBI melting model, an electric arc model and a coherent jet model. The simulation conditions were selected to match those typically seen in industry-scale EAFs
Topological Tropology of V.S. Naipaul’s Islamic Travelogues and Daniel Pipes’ Islamic History: Ahistorical Historicism
Get PDFNobel laureate V. S. Naipaul’s (1932-2018) first Islamic travelogue Among the Believers: An Islamic Journey (1981) contains his experience of a visit from August 1979 to February 1980 to the four non-Arab Muslim-majority countries – Iran, Pakistan, Malaysia, and Indonesia. Similarly, his last Islamic travelogue Beyond Belief: Islamic Excursions among the Converted Peoples (1998) has a description of another visit to the same countries for five-month in 1995. Concurrently, Daniel Pipes (1949-), an American historian, published his doctoral dissertation, Slave Soldiers and Islam: The Genesis of a Military System (1981), which represents Islamic culture as the first instigator of military slavery in the world. Then, he wrote an analysis of modern Islamic history In the Path of God: Islam and Political Power (1983), which historicizes Islam as a politically failed force all over the world. These travelogues and history are generically different. But a common topological relationality can be mapped in the anecdotes of Naipaul’s travelogues and the historiography of Pipes’ history, as they use identical tropological configurations to historicize Islamic cultures. This similar tropological historiography, this article argues, is covertly an offshoot of the contemporary spatiotemporal context in which they were produced. The context was networked by certain ideological implications, ethnocentrism, and some cultural misapprehensions regarding Islamic/Muslim culture, making the historicism of both Naipaul and Pipes seem ahistorical
Review of Fighting Invisibility: Asian Americans in the Midwest. Rutgers University Press
Get PDFDespite the imperative to shed light on the experiences of Asian Americans residing in the Midwest, previous research has predominantly centered around Asians residing on the coasts. This phenomenon, as elucidated by Erika Lee in her concept of “doubly invisible” Midwest Asian Americans, highlights their dual marginalization—they are not only conspicuously absent from mainstream literature but also from the broader discourse on Asian American experiences, which primarily revolves around those on the East and West coasts. However, it is within the heartland of America that we witness one of the most profound transformations in Asian America over the past few decades. Monica Trieu’s book, Fighting Invisibility, addresses this critical gap by drawing upon a comprehensive dataset, including 52 in-depth interviews with 1.5 to 2nd generation Asian Americans residing in the Midwest, census data, and cultural productions from Asian Americans residing in ten Midwestern states. This interdisciplinary research rigorously dissects the demographic landscape of Asian America in the Midwest, unveiling three prominent themes: spatially defined isolation, pervasive invisibility, and racialized visibility
Image Analysis for Plant Phenotyping
No full textPlant phenotyping focuses on the measurement of plant characteristics throughout the growing season, typically with the goal of evaluating genotypes for plant breeding and management practices related to nutrient applications. Estimating plant characteristics is important for finding the relationship between the plant’s genetic data and observable traits, which is also related to the environment and management practices. Recent machine learning approaches provide promising capabilities for high-throughput plant phenotyping using images. In this thesis, we focus on estimating plant traits for a field-based crop using images captured by Unmanned Aerial Vehicles (UAVs). We propose a method for estimating plant centers by transferring an existing model to a new scenario using limited ground truth data. We describe the use of transfer learning using a model fine-tuned for a single field or a single type of plant on a varied set of similar crops and fields. We introduce a method for rapidly counting panicles using images acquired by UAVs. We evaluate three different deep neural network structures for panicle counting and location. We propose a method for sorghum flowering time estimation using multi-temporal panicle counting. We present an approach that uses synthetic training images from generative adversarial networks for data augmentation to enhance the performance of sorghum panicle detection and counting. We reduce the amount of training data for sorghum panicle detection via semi-supervised learning. We create synthetic sorghum and maize images using diffusion models. We propose a method for tomato plant segmentation by color correction and color space conversion. We also introduce the methods for detecting and classifying bacterial tomato wilting from images