UARK (University of Arkansas )
Not a member yet
19829 research outputs found
Sort by
Towards Robust and Fair Vision Learning in Open-World Environments
Get PDFThe rapid increase of large-scale data and high-performance computational hardware has promoted the development of data-driven machine vision approaches. Advanced deep learning approaches have achieved remarkable performance in various vision problems and are closing the capability gap between artificial intelligence (AI) and humans. However, towards the ultimate goal of AI, which replicates human ability in visual perception tasks, the machine vision learning methods still need to address several ill-posed challenges. First, while the current vision learning methods often rely on large-scale annotated data, the data annotation process is a costly and time-consuming process. Second, the unfaired predictions produced by vision models due to the imbalance of data distribution, known as fairness, pose a significant concern in practical deployment, especially in human-related applications. Third, since human perceptions interpret the world in the open-vocabulary approach with diverse categories and concepts, the current vision machine frameworks should be capable of continually learning new concepts. Fourth, although the current deep learning-based vision approaches achieved impressive performance, their knowledge representations are often uninterpretable. These challenges motivate this dissertation to develop novel approaches toward fairness and robustness in vision learning. To address these challenges, the dissertation presents four key contributions toward fairness and robustness in vision learning. First, to address the problem of large-scale data requirements, the dissertation presents a novel Fairness Domain Adaptation approach derived from two major research findings. In particular, the thesis proposes a novel Bijective Maximum Likelihood to Unsupervised Domain Adaptation followed by introducing a novel Fairness Adaptation Learning Framework. Second, to enable the capability of open-world modeling of vision learning, this dissertation presents a novel Open-world Fairness Continual Learning Framework. The success of this research direction is the result of two research lines, i.e., Fairness Continual Learning and Open-world Continual Learning. Third, since visual data are often captured from multiple camera views, robust vision learning methods should be capable of modeling invariant features across views. To achieve this desired goal, the research in this thesis will present a novel Geometry-based Cross-view Adaptation framework to learn robust feature representations across views. Finally, with the recent increase in large-scale videos and multimodal data, understanding the feature representations and improving the robustness of large-scale visual foundation models is critical. Therefore, this thesis will present novel Transformer-based approaches to improve the robust feature representations against multimodal and temporal data. By introducing new self-attention mechanisms and learning objectives to Transformer networks for multimodal and video understanding, this research line has provided a comprehensive and better understanding of multimodal and temporal feature representations. Then, I will present a novel Domain Generalization Approach to improve the robustness of visual foundation models. My research\u27s theoretical analysis and experimental results have shown the effectiveness of the proposed approaches, demonstrating their superior performance compared to prior studies. I am confident that the contributions in this dissertation have advanced the fairness and robustness of machine vision learning
Design and Optimization of Matrix Core Transformer for High Power Bidirectional Isolated DC/DC Converter
Get PDFThanks to the advancement of semiconductor technology, the silicon carbide (SiC) devices with high voltage and power capability are developed and widely used in several industry applications, e.g., gird-connected photovoltaic (PV) system, data center power supply, and electrical vehicle (EV) charging station. After the concept of solid-state transformer (SST) was proposed as an important solution to the medium voltage (MV) power conversion systems, this bidirectional isolated DC/DC converter has attracted much attention in academia and industry. One of the main advantages of SST is to improve both the power density and efficiency of the power conversion system by replacing the bulky line-frequency transformer (LFT) with the medium-frequency transformer (MFT). In addition, SST has proved to have high voltage and power regulation capabilities with proper control methods. As a crucial component of SST, the MFT can significantly influence the performance of the power conversion system. The MFT represents a large proportion of the entire power conversion system in both loss and size aspects, and its temperature rise has a direct impact on reliability. Even though some research has discussed the design of MFT in SST, there still exist many challenges in the process with more and more stringent requirements. To achieve high power density, efficiency, and thermal capability, the MFT should be designed based on detailed analysis of SST operation principle, accurate electromagnetic thermal coupling models of MFT, and advanced optimization algorithms. In addition, some new structures of MFT can be proposed with different thermal and isolation considerations. In this work, the matrix core transformer (MCT) is presented and analyzed for high power bidirectional isolated DC/DC converters to address the aforementioned issues in MFT design. Combined with the operation principle of the power converter, the loss, leakage inductance, and thermal network models of MCT are proposed. With the above electromagnetic thermal coupling models, Pareto optimization and the nondominated sorting genetic algorithm III (NSGA-III) are introduced as the optimization engines for MCT design. Two high-power-density MCT prototypes (100 kW/50 kHz and 100 kW/100 kHz) for different applications are designed based on the optimization results and built with additive manufactured bobbin designs, which are designed to control the leakage inductance and provide airflow path. With both finite element analysis (FEA) simulation and experimental studies, the performances of optimal MCT designs are validated, especially including excellent heat dissipation capability
Deep Learning Framework for Inverse Problems in Computational Imaging: A Lensless Imaging and Super-resolution Magnetic Resonance Imaging Case
Get PDFInverse problems in computer vision involve reconstructing an original scene or image from incomplete, noisy, or indirect measurements. These problems are critical in tasks such as image denoising, deblurring, super-resolution, and lensless imaging, where the goal is to recover high-quality images from degraded or partial measurements. This thesis introduces novel approaches to address two specific real-world inverse problems: (1) image super-resolution in medical imaging and (2) lensless image reconstruction . In the first part of this work, we tackle the problem of image super-resolution in Magnetic Resonance Imaging (MRI). High-resolution MRI scans are often limited by hardware constraints, patient movement, and lengthy acquisition times. To address these challenges, we propose a fully unsupervised approach based on score-based diffusion models to super-resolve low-resolution MRI images. Unlike traditional methods that rely on simulated paired data, our model learns the underlying data distribution directly and reconstructs high-resolution MRI images from low-resolution inputs. Our method not only surpasses state-of-the-art supervised models in performance but also achieves faster sampling rates than current generative models for solving inverse problems. Additionally, we present an open dataset for training and benchmarking MRI super-resolution methods. In the second part, we propose an attention-based hybrid deep learning model for mask-based lensless imaging. Lensless imaging eliminates the need for expensive and bulky lenses, but traditional reconstruction algorithms suffer from slow convergence and poor perceptual image quality, while deep learning methods often introduce artifacts due to a lack of prior knowledge about the imaging model. Our approach integrates a conventional model-based optimization algorithm with an attention-based deep learning model. Experimental results demonstrate a significant improvement in perceptual quality compared to state-of-the-art methods
Structural Projections to the Nucleus Accumbens Link to Impulsive Components of Human Risk Preference
Get PDFFunctional responses in the Nucleus Accumbens (NAcc) to risk- and reward-related cues can predict real-life risk-taking behavior. Since NAcc activity depends on neurotransmission from connected brain regions, projections to the NAcc may also predict risk preference. To quantify risk preference, we employed latent variables previously derived in a comprehensive, independent study examining the psychometric structure of risk preference, which yielded a general risk preference factor as well as several specific factors, including a factor capturing impulsivity. Informed by previous work, we preregistered a set of hypotheses concerning the association between different risk preference factors and fractional anisotropy (or FA, which is sensitive to fiber coherence) for projections to the NAcc from Medial PreFrontal Cortex (MPFC), Anterior Insula, Amygdala, and an inferior tract from the Ventral Tegmental Area (iVTA). We tested our hypotheses in a community sample of 125 healthy human adults. As predicted, bilateral iVTA-NAcc tract FA showed a negative correlation with a psychometric factor that captured impulsivity, generalizing findings from prior research. Also as predicted, FA of the bilateral Amygdala-NAcc tract was positively associated with the impulsivity factor. Contrary to predictions, however, we observed no robust associations between the general risk preference factor and FA for projections from bilateral MPFC, right Anterior Insula, or bilateral Amygdala to the NAcc. Notably, exploratory unilateral analyses revealed an association between the general risk preference factor and left MPFC-NAcc tract FA. Taken together, these findings suggest that impulse control as a facet of risk preference maps onto specific neurobiological targets, while more general facets of risk preference may be supported by structural properties of lateral fronto-striatal projections. Although the exact associated functional mechanisms remain to be fully clarified, conNAcctomic approaches like the one presented here could pave the way for further research into the physiological foundations of risk preference and related constructs
Considering the Lessons of Curriculum Studies in the Design of Science Instruction: Varieties of Meaning and Implications for Teaching and Learning
Get PDFThis article is an introduction to the rich domain of curriculum studies with specific reference to science teaching and learning. It is designed not as a systematic review or theoretical treatise but rather as an overview for those charged with transforming science content and processes into a classroom curriculum. In other words, while written from the scholarly perspective of curriculum studies, the hope is that it will be seen by teachers as a set of possibilities rather than recommendations while reminding educators that what happens in classrooms to students and their parents is the curriculum, but it is only one of many that might have been delivered. To accomplish this, this paper explores the complex definition of curriculum including the notions of the kinds of curriculum from null, to formal, received, and learned with an emphasis on what occurs as a specific curriculum design results in effective and even faulty learning, a unique consequence proposed here. Next, we explore the common curriculum ideologies or orientations including those focused on academic advancement, tradition, student-centeredness, and social improvement. Finally, a formal recommendation for the content of science instruction in the U.S.—the Next-Generation Science Standards, is are considered as a conclusion by applying the expansive perspective of the term and nature of curriculum discussed throughout
Learning from Oversampling: A Systematic Exploitation of Oversampling to Address Data Scarcity Issues in Deep Learning- Based Magnetic Resonance Image Reconstruction
Get PDFData acquisitions in Magnetic Resonance Imaging (MRI) are inherently slow due to sequential acquisition protocol. Image reconstruction from under-sampled data is posed as an inverse problem in traditional model-based learning paradigms. Recent data-centric learning frameworks such as deep learning (DL) frameworks are data hungry, and demand a large, labeled training data sets. To address the lack of large training datasets, in MRI reconstructions, researchers approach the problem in two ways: (1) unsupervised method where the model is trained without the presence of fully sampled data. (2) using a method that efficiently use the limited dataset for training purpose. In this paper, we first systematically investigate advantages and limitations of current oversampling methods. Then, we also propose a novel oversampling method and a DL framework that systematically exploits the oversampling technique in the learning process as well as also increase the size of training data set. Essentially, we pose the training data oversampling as a one-to-many mapping function and introduce a new loss function based on similarity metric that can be integrated into a DL framework. Our proposed method not only addresses the training data scarcity in MR image reconstruction and improves reconstruction, but also makes the learned model more robust to different under-sampling techniques
Physiochemical, Bio, Thermal, and Non-Thermal Processing of Major and Minor Millets: A Comprehensive Review on Antinutritional and Antioxidant Properties
Get PDFMillets are recognized as future foods due to their abundant nutrition and resilience, increasing their value on the global stage. Millets possess a broad spectrum of nutrients, antinutrients, and antioxidants, making it imperative to understand the effects of various processing methods on these components. Antinutritional factors interfere with the digestibility of macro-nutrients and the bioavailability and bio accessibility of minerals. This necessitates methods to reduce or eliminate antinutrients while improving nutritive and antioxidant value in food. This review aims to elucidate the rationale behind processing choices by evaluating the scientific literature and examining the mechanisms of processing methods, categorized as physiochemical, bio, thermal, novel non-thermal, and their combination techniques. Physiochemical and bioprocessing methods alter antinutrients and antioxidant profiles through mass transfer, enzyme activation, product synthesis, microbial activity, and selective removal of grain layers. Thermal methods break functional bonds, modify the chemical or physical structures, enhance kinetics, or degrade heat-labile components. Non-thermal techniques preserve heat-sensitive antioxidants while reducing antinutrients through structural modifications, oxidation by ROS, and break down the covalent and non-covalent bonds, resulting in degradation of compounds. To maximize the trade-off between retention of beneficial components and reducing detrimental ones, exploring the synergy of combination techniques is crucial. Beyond mitigating antinutrients, these processing methods also stimulate the release of bioactive compounds, including phenolics, flavonoids, and peptides, which exhibit potent health-promoting properties. This review underscores the transformative potential of processing technologies in enhancing millets as functional ingredients in modern diets, promoting health and advancing sustainable food practices
An Evaluation of Bacterial Wilt (\u3ci\u3eRalstonia solanacearum\u3c/i\u3e) Resistance in a Set of Tomato Germplasm from the United States Department of Agriculture
Get PDFBacterial wilt (BW), caused by Ralstonia solanacearum, is one of the devastating diseases in tomatoes (Solanum lycopersicum L.). The use of resistant cultivars and breeding for genetic resistance is the most effective, economical, and environmentally friendly management strategy for this disease. It is necessary to screen diverse germplasm and cultivated genotypes to identify resistant resources and to develop resistant cultivars in tomatoes to combat the changing pathogen isolates. This study evaluated 40 United States Department of Agriculture (USDA) tomato accessions for their BW resistance to the R. solanacearum isolate P822 under greenhouse conditions. The tomato plants were inoculated and visually assessed to observe their symptoms, and the disease severity was scored on a scale of 0 to 4 (0 = no leaf wilted, 1 = 25% of leaves wilted, 2 = 50% leaves wilted, 3 = 75% of leaves wilted, and 4 = 100% leaves wilted). Five accessions (PI 645370, PI 647306, PI 600993, PI 355110, and PI 270210) were observed as BW resistance, with PI 645370 showing the greatest resistance. The broad-sense heritability for BW resistance was estimated as 59.9% and 42.8% based on a 0–4 scale of disease incidence and the disease severity index, respectively. Two distinct clusters (sub-populations) were detected among 39 of the 40 accessions. The five identified BW-resistant accessions were distributed in both clusters, suggesting a likely difference in the genetic base among the five resistance accessions. The resistant accessions will contribute significantly to the tomato breeding program to develop new cultivars with BW resistance
A Liquid Crystal-Based Biomaterial Platform for Rapid Sensing of Heat Stress Using Machine Learning
Get PDFNovel biomaterials that bridge the knowledge gap in coupling molecular/protein signatures of disease/stress with rapid readouts are a critical need of society. One such scenario is an imbalance between bodily heat production and heat dissipation which leads to heat stress in organisms. In addition to diminished animal well-being, heat stress is detrimental to the poultry industry as poultry entails fast growth and high yields, resulting in greater metabolic activity and higher body heat production. When stressed, cells overexpress heat shock proteins (such as HSP70, a well-established intracellular stress indicator) and may undergo changes in their mechanical properties. Liquid crystals (LCs, fluids with orientational order) are facile sensors as they can readily transduce chemical signals to easily observable optical responses. In this work, we introduce a hybrid LC–cell biomaterial within which the difference in the expression of HSP70 is linked to optical changes in the response pattern via the use of convolutional neural networks (CNNs). The machine-learning (ML) models were trained on hundreds of such LC-response micrographs of chicken red blood cells with and without heat stress. The trained models exhibited remarkable accuracy of up to 99% on detecting the presence of heat stress in unseen microscopy samples. We also show that cross-linking chicken and human RBCs using glutaraldehyde in order to simulate a diseased cell was an efficient strategy for planning, building, training, and evaluating ML models. Overall, our efforts build towards designing biomaterials that can rapidly detect disease in organisms that is accompanied by a distinct change in the mechanical properties of cells. We aim to eventuate CNN-enabled LC-sensors that can rapidly report the presence of disease in scenarios where human judgment could be prohibitively difficult or slow
Draft Genome Sequence of \u3ci\u3eRuoffia tabacinasalis\u3c/i\u3e Isolated from a Bovine Nasal Swab: A Novel Member of the Bovine Nasal Microbiota
Get PDFWe report the isolation and draft genome sequence of Ruoffia tabacinasalis, a novel member of the bovine nasal microbiota. The genome, which is estimated to be 90.5% complete, is composed of one contig comprising 2,363,349 bp with a GC content of 36.66%