136879 research outputs found
Sort by
Discovering Global False Negatives on the Fly for Self-supervised Contrastive Learning
No full textIn self-supervised contrastive learning, negative pairs are typically constructed using an anchor image and a sample drawn from the entire dataset, excluding the anchor. However, this approach can result in the creation of negative pairs with similar semantics, referred to as ���false negatives���, leading to their embeddings being falsely pushed apart. To address this issue, we introduce GLOFND, an optimization-based approach that automatically learns on the fly the threshold for each anchor data to identify its false negatives during training. In contrast to previous methods for false negative discovery, our approach globally detects false negatives across the entire dataset rather than locally within the mini-batch. Moreover, the per-iteration computation cost of our approach remains independent of the dataset size. Experimental results on image and image-text data demonstrate the effectiveness of the proposed method
Analytic and Semi-Analytic Calculations for Color Glass in the Weak Field Limit
No full textThe classical field approximation for color glass condensate can be solved using numerical methods and recursive analytic series solutions. The recursive analytic solution is known to have a resummation in Fourier space in the so called weak field limit. Based in this limit, the proper spacetime dependence of quantities related to the gluon two point function are derived. Namely, the gluon energy momentum tensor and corresponding angular momentum tensor, the initial motion of the nuclei after the collision, and the isotropic momentum broadening coefficient are computed.
The McLerran-Venugopalan model is one realization of color glass condensate to which our calculations can be applied. In many cases it can be used to obtain closed analytic expressions, however they are susceptible to UV and IR divergences. An alternative model which maintains UV regularity and softens the IR divergence is thus proposed. The UV region in this new model is regulated by accounting for local charge correlations in the transverse plane. The leading IR divergence is cured by enforcing global color neutrality. Analytic expressions are obtainable in this new model in the form of infinite series.
Novel insights from this work include an analytic understanding of the full time evolution of important physical quantities like energy density, pressure, and momentum broadening, including their late-time behavior; a scheme to systematically include non-constant color charge densities needed to compute realistic nuclei; a numerical understanding of the deceleration of nuclei in collisions, and (for the first time) a calculation of the transverse motion of nuclei
Evidence for Metabolism of Creatine by the Conceptus, Placenta, and Uterus for Production of ATP During Conceptus Development in Pigs
No full textDuring gestation in pigs, most embryonic mortality occurs during two gestational time points with high energy demands due to extensive cell proliferation and migration. Between Days 14 and 25, free-floating conceptuses (embryo/fetus and associated placental membranes) elongate and attach to the uterus. Between Days 50 and 70, the uterine-placental interface undergoes extensive folding and develops mature areolae to maximize support for development of the fetuses. We hypothesize that insufficient energy in the form of ATP in conceptuses and uterine tissue may contribute to conceptus loss in pigs. Creatine, an organic compound commonly stored in the muscle as phosphocreatine, can regenerate ATP through the creatine (Cr)-creatine kinase (CK)- phosphocreatine (PCr) pathway. However, the expression of factors involved in creatine metabolism has not been examined in conceptus and uterine tissues throughout gestation in pigs. In the present study, we performed real-time qPCR to quantify expression of mRNAs for enzymes and the creatine transporter involved in the creatine metabolic pathway in conceptus and uterine tissues from Days 10, 12, 15, 18, 20, 24, 30, 40, 60, and 90 of gestation. Results of qPCR analyses revealed increases in expression of AGAT, GAMT, CKM, CKB, and SLC6A8 mRNAs in conceptuses on Day 15, and a further increase in AGAT mRNA in the chorioallantois on Day 90 of gestation. Immunofluorescence staining of the uterine-placental interface from Days 15, 16, 20, and 25 corroborated qPCR results, with the expression of GAMT, CKM, and CKB proteins appearing to increase in conceptus Tr cells on Day 15. The presence of GAMT, CKB, and CKM proteins were confirmed with Western blot analyses. Levels of endometrial AGAT and CKM mRNAs increased on Day 15, CKB mRNA increased again on Day 30, and AGAT, GAMT, and SLC6A8 mRNAs increased significantly on Days 40, 60, and 90. Furthermore, HPLC analyses confirmed the presence of Cr and PCr metabolites in uterine luminal fluid, allantoic fluid, and amniotic fluid with significant increases in the uterine fluid and allantoic fluid on Days 11 and 40, respectively. Collectively, results of this study indicate that the Cr-CK-PCr pathway could establish sufficient energy stores to support cell proliferation and migration required for conceptus elongation, implantation, and remodeling of the uterine-placental interface during gestation in pigs
Teacher Motivations for Field Trips to Small Museums
No full textThis study explored what components are most valued by teachers during the planning phase of field trips to small museums. It identified teachers' motivations and constraints when planning a field trip to a small museum. In addition, it examined the influence of motivations and constraints on the likelihood of teachers taking a field trip and teachers��� willingness to put in effort to take a field trip after the planning phase. Past research on teacher motivations for field trips, however, reveals a need to improve in the area of small museums. Small museums are significant contributors to society by providing local access to education, creating job opportunities, promoting sustainability, and more. This study sought to find teachers' motives and constraints during field trip planning and the intention of taking a field trip after planning through survey data collected from teachers ranging from kindergarten to twelfth grade. The data analyzed were grouped into motives, constraints, likelihood, and willingness. The findings offer insight to small museums to make the necessary investments in school field trips that may increase revenue, enhance community engagement, and create better educational experiences
Connecting the Dots: Improving Information Extraction by Modeling the Non-Sequential Dependencies of Entity Mentions
No full textInformation Extraction (IE) aims at automatically extracting structured information from unstructured and semi-structured documents. It is an important and challenging topic in Natural Language Processing (NLP) that plays a critical role in downstream applications such as Question Answering and Summarizing. It includes many sub-tasks like Named Entity Recognition (NER), Relation Extraction, Event Extraction, Table Annotations, etc.
Previous works on IE mainly focus on extracting knowledge from a document sentence by sentence and the text encoding models (e.g., RNNs, CNNs, and Transformers) regard the text as a linear sequence from the left to the right or from the right to the left. However, we humans do not always understand a document by sequentially reading it. We tend to connect the concepts across the whole document and then form structural knowledge in our brains. Motivated by the intuition, this work proposes to introduce the non-sequential connections within a document, and uses the relations among the entity mentions as the surrogates for the non-sequential conceptual connections to further improve the performance of the IE systems.
Firstly, I propose to connect the related entity mentions in a document and enforce information flow among them for consistent entity type predictions. Specifically, the work connects both the local dependency relations and global coreference relations for the entity mentions to build better entity mention representations. Experimental results show that applying Graph Neural Networks (GNNs) on the connections can improve the NER performance over strong baselines on two domain-specific datasets.
Secondly, I propose to connect the conceptually related regions in a document and encourage semantic interactions within and among regions. In particular, it builds the connections among the candidate role fillers (i.e., entity mentions from an event mention) for the event extraction task, and characterizes the connections based on different regional affiliations. Then edge-aware GNNs are applied to update the representations of the candidates for false positive filtering. Empirical results show that the proposed method can yield new state-of-the-art performance on two document-level event extraction datasets in two different languages.
Lastly, I propose to connect the entity mentions from semi-structured tables and incorporate the table structures into the table element representations, benefiting table annotation (knowledge extraction from tables) tasks. Specifically, the system will first build hyper-graphs for the cell values coming from the same row or column and then apply the hyper-graph Neural Networks to learn better table representations. The evaluation and analysis show the effectiveness of the structure-aware table representations in improving the table annotation tasks
Geometric Deep Learning for Molecular Discoveries
No full textWith the rapid advancement of artificial intelligence (AI), its applications in scientific research have grown significantly, giving rise to the research area of AI for science (AI4Science). In this dissertation, we focus on AI for molecular science, such as small molecules and proteins, aiming to build efficient and effective methods to accelerate molecular discovery. we particularly focus on two fundamental tasks, molecular representation learning and molecule generation. Specifically, we model molecules as graphs and design geometric deep learning methods for molecules.
We first consider representation learning of 3D molecular graphs, where each node has its 3D coordinates. With an accurate representation learning model, we can reduce the computation time required for predicting molecular properties. In this dissertation, we provide an analysis in the spherical Coordinate System (SCS) for the complete identification of 3D graph structures and propose our SphereNet. SphereNet can distinguish similar molecular structures, such as two enantiomers that are mirror images of each other and reduce complexity from O(nk��) to O(nk��),
enabling it to perform efficiently on large-scale molecules. Here n and k denote the number of nodes and the average degree in the 3D graph, respectively.
While SphereNet presents advancements in accuracy and efficiency, it still can not incorporate 3D information completely. Furthermore, its complexity remains higher than some existing methods. We then propose ComENet to address these issues and incorporate 3D information completely and efficiently. Our method guarantees full completeness of 3D information on 3D graphs by achieving global and local completeness with a complexity of O(nk).
SphereNet and ComENet are tailored for small molecules. Extending their application to proteins is challenging due to the large number of atoms in proteins and their inherent multi-level nature. Therefore, we further design our method ProNet specifically for proteins. ProNet completely captures three levels of protein structures, e.g., the amino acid, backbone, or all-atom levels and is more efficient than existing methods. ProNet can be applied on various downstream tasks, including protein fold and function prediction, protein-ligand binding affinity prediction, and protein-protein interaction prediction.
Lastly, we consider 3D molecule generation. The generation of novel molecules with desired properties is an important step in drug discovery. In this dissertation, we apply language models (LMs) for 3D molecule generation by introducing our canonical and SE(3)-invariant tokenizer, Geo2Seq. Experiments show that our new method can achieve promising results
Crowding: An Exploration of the Effects of Methods and Sound
No full textVisitation rates continue to rise in the United States��� protected areas like national parks and national forests. This has raised management concerns for both impacts on the environment and visitor experiences. The Satisfaction Model postulates that as use levels and encounters rise in parks and protected areas, there is a threshold where the visitor experience is negatively impacted by the additional visitors. Nevertheless, decades of research indicate this relationship is complicated and multifaceted. As a result, the Satisfaction Model has evolved to include norms, use patterns and research measurement techniques as concepts that impact responses to encountering other individuals. Therefore, the purpose of this dissertation was to explore how people respond to research techniques and environment conditions, specifically the soundscape, when visiting protected areas.
Study one focused on starting point bias; a research bias where participant responses are systematically inflated or deflated due to research techniques. Research on crowding responses has frequently relied on visual methods where participants are shown a series of images with varying numbers of people visiting a protected area. The order in which the images are shown holds the potential to inflate or deflate results because the exposure to one treatment may impact responses to a subsequent treatment. This study���s findings revealed a starting point bias, but only on crowding ratings when moving from low setting density to high setting density.
Furthermore, the soundscape has become a fruitful topic for research on visitor experiences. However, little is understood about how sound impacts crowding norms. Study two explored how anthropogenic sound types impact crowding and acceptability ratings of a setting density. Direct human sounds like voices and children playing were expected to be rated more favorably than mechanical sounds. However, the rank of crowding and acceptability ratings were mixed with direct human sounds generally being rated more harshly than mechanical sounds.
Study three explored how sound loudness impacted crowding and acceptability ratings of the setting. Results indicated that loudness was only rated more harshly at the highest loudness levels. The studies are discussed in terms of their results and their impacts on theory and practice
Characterization of Ligand Binding Using Dissolution DNP Assisted NMR Spectroscopy
No full textBiomolecular interactions play essential roles in cellular processes including signaling, metabolism, and enzymatic synthesis of cellular components. Elucidating interactions between proteins and ligands using techniques such as nuclear magnetic resonance (NMR) spectroscopy provides fundamental insights into biological function, as well as guidance on the identification of new drug candidates. A significant NMR sensitivity improvement of several thousand-fold can be achieved by hyperpolarizing the ligand molecule using dissolution dynamic nuclear polarization (D-DNP). Spectra can be acquired in a reduced time, at or near physiological concentrations. Here, transverse (R2) NMR relaxometry is demonstrated to probe protein-ligand interactions. A 13C R2 relaxation dispersion measurement characterizes ligand binding epitopes through the observation of relaxation rates at different positions of the ligand, whereby the magnitude of the dispersion reflects the binding orientation of the ligand. The efficiency of the R2 measurement can be improved by an ultrafast approach to obtain the relaxation rates from all 13C spins in a single measurement. Numerous target proteins for pharmaceuticals are embedded in the cell membrane. Hyperpolarized 19F, due to its low NMR detection limit, is proposed for probing the interactions with membranes and the cell surface proteins. A model for the binding interaction combined with predictions of spin relaxation rates provides estimates of the binding affinity to membranes of different compositions in unilamellar vesicles. Applied to the measurement of ligand interactions with different cell types, the influence of the interactions between ligands and cell membrane proteins is identified
A Comparative Analysis of Archaeological and Textual Evidence for Danubian Watercraft, First to Fourth Centuries CE
No full textThe Danube is one of the longest rivers in the world and has been a life source for peoples settled around it for millennia. Alongside local groups, Greeks and Romans used the river as a throughway. Though there are many accounts of the Danube in Roman history, few of them provide insight into what types of vessels were using along the river, despite many documented interactions with local groups. The archaeological record along the Lower Danube, that is the section from the Iron Gates to the Danube Delta, has begun to reveal more evidence of both Roman and local boats used along the ancient course of the Danube, but many of these are found in rescue excavations with limited recording available. In this thesis I combine these two fragmentary sources to examine if the literary accounts are reflective of the archaeological evidence regarding local boatbuilding traditions. Though brief, there are terms in literature which indicate construction methods (����������������), and hull shape (cavatis arborum alveis vs. truncis), which are reflected several times in the archaeological record. Other literary descriptions reveal little about the vessels themselves, but the contexts in which they are found are reflective of Roman perceptions of the users of the vessels. These sources both undermine the capability of the locals to boost military morale and also demonstrate their efficacy, as the Romans were intimidated by those who used the Danube for raiding and stealth attacks. Flat-bottomed barges fastened with closely spaced iron clamps also appear in the archaeological record several times along the Lower Danube and in the greater Danube Basin. Further archaeological research is needed to bolster this theory of a regional vessel type, and such work may continue to yield multiple ship types in the same contexts, increasing the number and variety of vessels from antiquity and perhaps revealing a regional tradition through time. Though there is still much to be excavated, this research has revealed the disparity between Roman descriptions of vessels in use along the Lower Danube and the reality of such boats by examining the archaeological evidence and contextualizing the Roman literature
Investigating Fifth Grade Teachers��� Implementation of ESL Instructional Strategies in Literacy-Infused Science Instruction
No full textThe purpose of this study was to investigate whether there is a significant difference in the time allocation of ESL instructional strategies between the treatment teachers who received literacy-infused science (LIS) curriculum accompanied virtual professional development (VPD) on such strategies as an intervention and control group of teachers who did not receive the VPD intervention. I further explored teachers��� perceptions of the impact and challenges of English as a Second Language (ESL) instructional strategy implementation in their science teaching practices and their students��� responses to the strategies.
To examine the differences in fifth-grade science teachers��� time allocation in their use of ESL instructional strategies between treatment and control groups, a low-inference observational instrument, Pedagogical Observation Protocol (POP), was used. Treatment teachers received bi-weekly VPD sessions provided through Project LISTO (Literacy-Infused Science Using Technology Innovation Opportunity, Grant Award No. U411B16001; Lara-Alecio et al., 2013), including a series of scaffolding strategies that benefit teachers��� professional growth as well as their students��� learning outcomes. A total of 14,332 rounds of observation clips were collected from 98 in-service science teachers who were randomly assigned to treatment and control conditions across Texas during the 2018-2019 academic year. Results revealed a statistically significant difference in 7 out of 9 strategies between treatment and control group teachers in utilizing ESL instructional strategies in science classrooms. Treatment teachers significantly used more strategies learned in VPDs than the control group of teachers. To further explore the perceived impact of ESL instructional strategy implementation, semi-structured focus group interviews were conducted for treatment teachers at the end of the school year. The results of the thematic analysis indicated a positive perception of using ESL strategies in science classrooms in terms of enhancing teachers��� quality of instruction and self-confidence as well as students��� science learning interest and literacy development. Therefore, this evidence supported that teachers��� instructional practice and student support were positively impacted by allocating a variety of instructional time toward ESL strategies. However, teachers encountered the challenges such as time constraints and technological issues while implementing ESL strategies might potentially lead to less time allocation in certain strategies