Sparse Data Research Articles

Image to Label to Answer: An Efficient Framework for Enhanced Clinical Applications in Medical Visual Question Answering

Medical Visual Question Answering (Med-VQA) faces significant limitations in application development due to sparse and challenging data acquisition. Existing approaches focus on multi-modal learning to equip models with medical image inference and natural language understanding, but this worsens data scarcity in Med-VQA, hindering clinical application and advancement. This paper proposes the ITLTA framework for Med-VQA, designed based on field requirements. ITLTA combines multi-label learning of medical images with the language understanding and reasoning capabilities of large language models (LLMs) to achieve zero-shot learning, meeting natural language module needs without end-to-end training. This approach reduces deployment costs and training data requirements, allowing LLMs to function as flexible, plug-and-play modules. To enhance multi-label classification accuracy, the framework uses external medical image data for pretraining, integrated with a joint feature and label attention mechanism. This configuration ensures robust performance and applicability, even with limited data. Additionally, the framework clarifies the decision-making process for visual labels and question prompts, enhancing the interpretability of Med-VQA. Validated on the VQA-Med 2019 dataset, our method demonstrates superior effectiveness compared to existing methods, confirming its outstanding performance for enhanced clinical applications.

DAPM-CDR: A domain adaptation prompting model for drug response prediction

The rising incidence and mortality rates of cancer present significant challenges to global health. Variations in tumor growth rates and treatment responses have revealed the limitations of traditional therapies, highlighting the urgent need for predictive models for cancer drug responses based on computational methods. Current drug response prediction methods often fall short in accurately predicting responses for rare cancers with limited data. To address these issues, we propose a domain adaptation prompting model for drug response prediction, DAPM-CDR. DAPM-CDR is trained with cancer types with rich response data of cancer cell lines, integrating both common and specific information into prompts through contrastive learning, to transfer knowledge to target tasks that with sparse data on cancer cell line responses, thereby enhancing the generalization capabilities across various cancer types. The experiment results demonstrate that DAPM-CDR outperforms several competitive methods in predicting drug responses of cell lines, particularly excelling with data from rare cancers and showing significant performance enhancements.

Intelligent Bayesian regularization backpropagation neuro computing paradigm for state features estimation of underwater passive object

In underwater environments, the accurate estimation of state features for passive object is a critical aspect of various applications, including underwater robotics, surveillance, and environmental monitoring. This study presents an innovative neuro computing approach for instantaneous state features reckoning of passive marine object following dynamic Markov chains. This paper introduces the potential of intelligent Bayesian regularization backpropagation neuro computing (IBRBNC) for the precise estimation of state features of underwater passive object. The proposed paradigm combines the power of artificial neural network with Bayesian regularization technique to address the challenges associated with noisy and limited underwater sensor data. The IBRBNC paradigm leverages deep neural networks with a focus on backpropagation to model complex relationships in the underwater environment. Furthermore, Bayesian regularization is introduced to incorporate prior knowledge and mitigate overfitting, enhancing the model’s robustness and generalization capabilities. This dual approach results in a highly adaptive and intelligent system capable of accurately estimating the state features of passive object in real-time. To evaluate the efficacy of this intelligent computing approach, a controlled supervised maneuvering trajectory for underwater passive object is constructed. Real-time estimations of location, velocity, and turn rate for dynamic target are scrutinized across five distinct scenarios by varying the Gaussian observed noise’s standard deviation, aiming to minimize mean square errors (MSEs) between real and estimated values. The effectiveness of the proposed IBRBNC paradigm is demonstrated through extensive simulations and experimental trials. Results showcase its superiority over traditional nonlinear filtering methods like interacting multiple model extended Kalman filter (IMMEKF) and interacting multiple model unscented Kalman filter (IMMUKF), especially in the presence of noise, incomplete measurements and sparse data.

Optical projection tomography implemented for accessibility and low cost (OPTImAL).

Optical projection tomography (OPT) is a three-dimensional mesoscopic imaging modality that can use absorption or fluorescence contrast, and is widely applied to fixed and live samples in the mm-cm scale. For fluorescence OPT, we present OPT implemented for accessibility and low cost, an open-source research-grade implementation of modular OPT hardware and software that has been designed to be widely accessible by using low-cost components, including light-emitting diode (LED) excitation and cooled complementary metal-oxide-semiconductor (CMOS) cameras. Both the hardware and software are modular and flexible in their implementation, enabling rapid switching between sample size scales and supporting compressive sensing to reconstruct images from undersampled sparse OPT data, e.g. to facilitate rapid imaging with low photobleaching/phototoxicity. We also explore a simple implementation of focal scanning OPT to achieve higher resolution, which entails the use of a fan-beam geometry reconstruction method to account for variation in magnification. This article is part of the Theo Murphy meeting issue 'Open, reproducible hardware for microscopy'.

A Two‐stage Bayesian Small Area Estimation Approach for Proportions

SummaryWith the rise in popularity of digital Atlases to communicate spatial variation, there is an increasing need for robust small area estimates. However, current small area estimation methods suffer from various modelling problems when data are very sparse or when estimates are required for areas with very small populations. These issues are particularly heightened when modelling proportions. Additionally, recent work has shown significant benefits in modelling at both the individual and area levels. We propose a two‐stage Bayesian hierarchical small area estimation approach for proportions that can account for survey design, reduce direct estimate instability and generate prevalence estimates for small areas with no survey data. Using a simulation study, we show that, compared with existing Bayesian small area estimation methods, our approach can provide optimal predictive performance (Bayesian mean relative root mean squared error, mean absolute relative bias and coverage) of proportions under a variety of data conditions, including very sparse and unstable data. To assess the model in practice, we compare modelled estimates of current smoking prevalence for 1,630 small areas in Australia using the 2017–2018 National Health Survey data combined with 2016 census data.

Effects of Phosphorylation on Protein Backbone Dynamics and Conformational Preferences.

Phosphorylations are the most common and extensively studied post-translational modification (PTM) of proteins in eukaryotes. They constitute a major regulatory mechanism, modulating protein function, protein-protein interactions, as well as subcellular localization. Phosphorylation sites are preferably located in intrinsically disordered regions and have been shown to trigger structural rearrangements and order-to-disorder transitions. They can therefore have a significant effect on protein backbone dynamics or conformation, but only sparse experimental data are available. To obtain a more general description of how and when phosphorylations have a significant effect on protein behavior, molecular dynamics (MD) currently provides the only suitable framework to study these effects at a large scale in atomistic detail. This study develops a systematic MD simulation framework to explore the influence of phosphorylations on the local backbone dynamics and conformational propensities of proteins. Through a series of glycine-backbone peptides, we studied the effects of amino acid residues including the three most common phosphorylations (Ser, Thr, and Tyr), on local backbone dynamics and conformational propensities. We further extended our study to investigate the interactions of all such residues between position i to positions i + 1, i + 2, i + 3, and i + 4 in such peptides. The final data set comprises structural ensembles for 3393 sequences with more than 1 μs of sampling for each ensemble. To validate the relevance of the results, the structural and conformational properties extracted from the MD simulations are compared to NMR data from the Biological Magnetic Resonance Data Bank. The systematic nature of this study enables the projection of the gained knowledge onto any phosphorylation site in the proteome and provides a general framework for the study of further PTMs. The full data set is publicly available, as a training and reference set.

ANAGL: A Noise-resistant and Anti-sparse Graph Learning for micro-video recommendation

In recent years, Graph Convolutional Networks (GCNs) have seen widespread utilization within micro-video recommendation systems, facilitating the understanding of user preferences through interactions with micro-videos. Despite the commendable performance exhibited by GCN-based methodologies, several persistent issues demand further scrutiny. Primarily, most user-micro-video interactions involve implicit behaviors, such as clicks or abstentions, which may inadvertently capture irrelevant micro-video content, thereby introducing significant noise (false touches, low watch-ratio, low ratings) into users’ histories. Consequently, this noise undermines the efficacy of micro-video recommendations. Moreover, the abundance of micro-videos has resulted in fewer interactions between users and micro-video content. To tackle these challenges, we propose a noise-resistant and anti-sparse graph learning framework for micro-video recommendation. Initially, we construct a denoiser that leverages implicit multi-attribute information (e.g., watch-ratio, timestamp, ratings, etc.) to filter noisy data from user interaction histories. This process yields high-fidelity micro-video information, enabling a more precise modeling of users’ feature preferences. Subsequently, we employ a multi-view reconstruction approach and utilize cross-view self-supervised learning to gain insights into user and micro-video features. This strategic approach effectively mitigates the issue of data sparsity. Extensive experiments conducted on two publicly available micro-video recommendation datasets validate the effectiveness of our proposed method. For in-depth details and access to the code, please refer to our repository at “ https://github.com/kbk12/ANAGL.git .”

CTITF: A tensor factorization model with constrained bidirectional user trust and implicit feedback for context-aware recommender systems

Recommender systems offer an efficient solution to the problem of information overload, which is exacerbated by the rapid expansion of data. Context-aware recommender systems (CARS) have become prominent, incorporating contextual information to provide more desirable recommendations. While the dimension of contextual information rises, the complexity of algorithms increases as well, highlighting the data sparsity problem. In this paper, we propose CTITF, a novel tensor factorization model that incorporates constrained user bidirectional trust and implicit feedback. First, CTITF distinguishes the roles of trust networks in terms of the directionality of trust relationships: the trustor and the trustee. CTITF then constructs trust influences and biases for both roles to linearly represent user preferences. Furthermore, CTITF improves the rating prediction function and recommendation objective function based on CP factorization. Finally, CTITF filters out trust relationships where users do not have the same history items. It also uses contextual information to limit the amount of implicit feedback, addressing the issue of high time complexity. Experiments on two real-world datasets illustrate that the proposed model outperforms previous TF-based models and social recommender systems in terms of rating prediction accuracy. In addition, the CTITF model reduces the training time by 50.28% compared to the trustTF model, which also uses trust.

Individual and team profiling to support theory of mind in artificial social intelligence.

We describe an approach aimed at helping artificial intelligence develop theory of mind of their human teammates to support team interactions. We show how this can be supported through the provision of quantifiable, machine-readable, a priori information about the human team members to an agent. We first show how our profiling approach can capture individual team member characteristic profiles that can be constructed from sparse data and provided to agents to support the development of artificial theory of mind. We then show how it captures features of team composition that may influence team performance. We document this through an experiment examining factors influencing the performance of ad-hoc teams executing a complex team coordination task when paired with an artificial social intelligence (ASI) teammate. We report the relationship between the individual and team characteristics and measures related to task performance and self-reported perceptions of the ASI. The results show that individual and emergent team profiles were able to characterize features of the team that predicted behavior and explain differences in perceptions of ASI. Further, the features of these profiles may interact differently when teams work with human versus ASI advisors. Most strikingly, our analyses showed that ASI advisors had a strong positive impact on low potential teams such that they improved the performance of those teams across mission outcome measures. We discuss these findings in the context of developing intelligent technologies capable of social cognition and engage in collaborative behaviors that improve team effectiveness.

Advancing precision medicine in cutaneous oncology through 3D high-frequency ultrasound imaging.

e13625 Background: Clinical assessment of tumor depth and shape often requires biopsy, especially when evaluating skin cancers. High Frequency Ultrasound (HFUS) is a non-invasive clinical tool that can quickly produce high resolution images ideally suited for evaluating topical tumors. In comparison to other imaging modalities, HFUS do not always provide the spatial data required to model 3-dimensional (3D) tumors. Currently, 3D-images are an amalgamation of many segmented image slices. However, in outpatient clinics it is unrealistic for HFUS devices and clinicians to gather large sets of image data. In our present original research, we aimed to overcome limited number of HFUS and dermatoscopic images by constructing a simple framework to process sparse and non-spatial data into working 3D skin tumor models. Methods: We analyzed two different skin tumors, the dermal duct tumor (DDT) and basal cell carcinoma (BCC), each with different number of HFUS images at 30Hz and 50Hz respectively. 1 set of 6 parallel images were obtained for the BCC, while 2 orthogonal sets of 6 parallel images were obtained for the DDT. For each tumor, a single dermatoscopic photo and sets of bitmap HFUS images with a resolution of 1024 x 512 were taken. Our study was IRB-approved. Mathematica and MATLAB, both with accessible documentation, were used to process images. Results: We developed a novel algorithmic approach to produce 3D tumor models with limited medical data. Both tumors appear corporally hypoechoic under HFUS. Mathematica was used to first reduce HFUS image noise and binarize the tumor region with a bilateral Gaussian filter. The user-friendly MATLAB Image Segmenter application manually filled the region of interest. Next, Mathematica was used to orient and plot tumor borders in 3D. Finally, we encompassed the borders with a 3D convex hull to conservatively predict tight margins around the tumor. Conclusions: Our simplified approach allows clinicians and scientists to produce 3D renderings of tumor margins. Margin control must always be balanced with a positive surgical outcome. By approaching our 3D-model with a convex hull around sampled border cross-sections, we chose a conservative approach that focuses skin tumor margin evaluation. Our results promote precision medicine and planning into the current standard of skin cancer treatment. Our findings can allow tracking tumor progression or treatment possible via non-invasive evaluation of tumor depth and contour. An aim of our present research is to lower the barrier of 3D construction in clinical settings. Simple algorithms, such as shown in our work, transform orthogonal and sparse data into usable data that can be implemented to train machine learning and artificial intelligent programs. Future research tailoring this approach may allow instant programming of HFUS images to non-invasively render cutaneous tumor borders and depth in real time during clinical examination.

FDM: Effective and efficient incident detection on sparse trajectory data

Incident detection (ID), or the automatic discovery of anomalies from road traffic data (e.g., road sensor and GPS data), enables emergency actions (e.g., rescuing injured people) to be carried out in a timely fashion. Existing ID solutions based on data mining or machine learning often rely on dense traffic data; for instance, sensors installed in highways provide frequent updates of road information. In this paper, we ask the question: can ID be performed on sparse traffic data (e.g., location data obtained from GPS devices equipped on vehicles)? As these data may not be enough to describe the state of the roads involved, they can undermine the effectiveness of existing ID solutions. To tackle this challenge, we borrow an important insight from the transportation area, which uses trajectories (i.e., moving histories of vehicles) to derive incident patterns. We study how to obtain incident patterns from trajectories and devise a new solution (called Filter-Discovery-Match (FDM)) to detect anomalies in sparse traffic data. We have also developed a fast algorithm to support FDM. Experiments on a taxi dataset in Hong Kong and a simulated dataset show that FDM is more effective than state-of-the-art ID solutions on sparse traffic data, and is also efficient.

Frequency of Hematologic and Nonhematologic Conditions in Pediatric Patients with Vascular Anomalies: A Pediatric Health Information System Database Study

Background: Vascular anomalies (VA) are a heterogeneous group of diseases which include vascular malformations and vascular tumors. In the pediatric population, there is a sparsity of data on the frequency of associated hematologic and nonhematologic conditions as well as overall outcomes for patients. Objectives: The objective of this study was to determine the rate of certain concurrent hematologic and nonhematologic conditions in patients with VA who underwent procedures and compare those to patients who did not undergo procedures. To examine the rates of intensive care unit admissions and discharge mortality in patients who had a vascular anomaly. Methods: This is a multicenter retrospective cohort study using data from the Pediatric Health Information System. The database was queried for children 0–18 years old with a diagnosis of a VA who had inpatient encounters between January 2016 and September 2021. Results: The most common hematologic conditions were thrombocytopenia, iron deficiency anemia, and venous thrombosis. The most common nonhematologic condition was electrolyte imbalances. Intensive care unit admissions were highest among patients with other congenital malformations of the peripheral vascular system, cerebral cavernous malformations, and sporadic arteriovenous malformations. Discharge mortality was highest in patients with lymphatic malformations and primary lymphedema. Conclusions: This is the largest database study of pediatric patients with VA and their concurrent medical conditions and outcomes to date. While our study could not prove associations, we hope that our results will serve as a foundation for more extensive research regarding the impact of these identified conditions and outcomes and contribute to improved management of patients with VA.

Testosterone replacement therapy (TRT) in patients with locoregional prostate cancer (LPC) treated with prior androgen deprivation therapy (ADT): A single center review.

e24045 Background: In LPC, ADT may be combined with radiation therapy (RT) for 4-24 months. Post-ADT, a significant proportion of men experience prolonged hypogonadism. TRT can alleviate symptoms and improve metabolic outcomes. Contrary to longstanding concerns, there is no substantial evidence of increased prostate cancer recurrence in patients (pts) treated with single modalities (surgery or RT). However, sparse data ( < 100 cases published to our knowledge) exist regarding recurrence risk of LPC with TRT after prior ADT, where baseline T levels may fall below the androgen receptor saturation point. Methods: We retrospectively abstracted clinical data from hospital records of men with stage I – IVA PC, treated with prior concurrent ADT and RT and subsequently received TRT between January 2014 and September 2023. Extracted data included demographics, cancer diagnosis/staging, ADT/TRT treatment details and PSA and clinical outcomes. The co-primary endpoints were change in PSA and incidence of PC recurrence. Results: 21 pts met criteria. Median age was 77. ISUP Grade Groups (GG) included: GG1: 2, GG2: 2, GG3: 8, GG4: 2, and GG5: 7. AJCC 8th ed. stages were: I: 2, II: 7, III: 7, IV-A: 5. The median duration of prior ADT was 8 months (IQR 5 – 17) and the median interval from RT to TRT was 19 months (IQR: 12 – 44). Prior to TRT, the median testosterone level was 30.5 ng/dL (IQR: 17-76). TRT formulations included: injection: 6, oral: 4, topical: 13. The median follow-up period from the start of TRT was 13 months, and the median duration on TRT was 10 months (IQR: 5-22). TRT was ongoing in 15 (71.4%) pts and discontinued in 6 (28.6%). Reasons for discontinuation included testosterone recovery (1), hospice (not PC related) (2), no perceived benefit (2), MD concern for PSA rise (1). After TRT, the median testosterone level was 336 ng/dL (IQR: 207-462). Median PSA pre- and post-TRT were undetectable and 0.08 ng/dL, respectively. None of the subjects experienced PC recurrence or PC-specific mortality. One pt showed PSA bounce without recurrence. Median BMI was unchanged before (28.3 kg/m2) and after (27.7 kg/m2) TRT (p = 0.48). Due to the nature of the study, quality of life measures and metabolic parameters could not be systematically abstracted. Conclusions: In men with LPC who remained hypogonadal long-term after prior ADT and RT, we found that TRT was not associated with a significant rise in median PSA and no cases of PC recurrence were documented. This included many with a history of very high-risk cancers (eg. GG ≥ 4, Stage ≥ III). This study adds to the sparse existing literature in this clinical setting and together they support the case for a prospective trial utilizing TRT in patients who remain hypogonadal after ADT and radiation.

Cluster-based Bayesian approach for noisy and sparse data: application to flow-state estimation

This study presents a cluster-based Bayesian methodology for state estimation under realistic conditions including noisy data from sparse sensors. The proposed approach is interpretable and, building upon previous work on transition networks, explicitly accounts for experimental noise within the data-driven framework by means of data clustering. Experimental measurements are exploited, beyond model training, to quantify the degree of uncertainty (noise) for each trained state. Such noise levels are eventually associated with probability distributions that, when combined with Bayes’ theorem, allow us to perform real-time state estimation. The proposed methodology is tested on two cases of challenging flows generated by an accelerating elliptical plate and also a delta wing experiencing gusts. Results specifically indicate that the proposed approach is robust against the number of clusters, enabling state estimation with a significant order reduction, notably decreasing the computational cost while preserving estimation accuracy. Based on the present findings, the proposed data-driven approach can be employed for realistic state estimation in nonlinear systems where noise, sensor sparsity and nonlinearities represent a challenging scenario.

Small bowel adenocarcinoma: A SEER database analysis.

e16383 Background: Small bowel malignancies (SBM) account for about 3% of all gastrointestinal malignancies. Small bowel adenocarcinoma (SBA) accounts for the majority of the SBM. SBA is associated with poor prognosis and treatment options are limited in patients with advanced cancer. We analyzed SEER data to study the factors that impacted survival in patients with SBA. Methods: We reviewed data on the SEER database (version Stat 8.4). We extracted data to include patients with adenocarcinoma with labeled primary site as small bowel diagnosed between 2000-2019. We excluded other histology. Variables included were age at diagnosis, race, gender, stage at diagnosis, and grade of malignancy. Chi-square test was used to analyze the association between each variable and survival. Results: We identified 3415 cases of SBA that met our inclusion criteria. The number of annual cases increased from 2000-2019. The mean age was 65.3 years (range 57-75 years). The Male to Female ratio was 11:9. 76% were white, 13.7% were black, 13% were others. Duodenum was the most common primary site with 58%, jejunum at 15.5%, ileum at 13%, and 13.5% overlapped sites or unknown. SEER combined summary stage at diagnosis was 32% regional, 18% localized, 29% distant, and 23% unknown. 6% was grade I, 37% grade II, 27% grade III, 1% grade IV, and 28% unknown. Median survival in patients with SBA was 13 months. The median survival for patients less than or equal to 60 years was 24 months compared to 11 months in patients over 60 years of age (p < 0.0001). Median survival in males was 13 months vs 12 months in females, (p = 0.548). There was no significant difference in survival per race (p = 0.5849). In this cohort, jejunal adenocarcinoma had better outcomes with a survival of 27 months (p < 0.0001). Grade I and II were associated with better outcomes with survival of 21.5 months and 24 months (p < 0.0001). Multivariate analysis showed that age less than 60, early-stage disease, jejunal primary site, and low-grade disease were associated with a better outcome. Conclusions: The number of cases of SBA in the SEER database has increased over the last few decades. SBA is associated with poor outcomes and our analysis shows that patients with early-stage disease, and younger age have better outcomes. Management of SBA is based on retrospective data and data extrapolated from management of colon adenocarcinoma, due to sparsity of prospective data. Clinical trials are needed to identify appropriate therapeutic options for the management of SBA.

Spatio-Temporal Hierarchical Data Structure Based on Sparse Voxel Octrees for Representation of the Geometry of Time-Variable Three-Dimensional Scenes

Abstract The paper deals with the problematics of representation of the geometry of three-dimensional scenes in computer graphics using domain-specific hierarchical data structures based on octant trees or directed acyclic graphs as the means of space saving binary representation of this information. Several hierarchical data structures were introduced in this field in past decades. In this paper we are moving this problematic to the field of the representation of three-dimensional scenes which are rather dynamic in time, so their geometry is changing on the regular basis. We are investigating the influence of this time-variability on the size of the related hierarchical data structure on binary level. We are comparing the approach in which each time step is considered as the stand-alone static three-dimensional scene and therefore is represented by its own hierarchical data structure with the approach when this time-variable scene is considered as the four dimensional space and spatio-temporal sparse voxel octree hierarchical data structure is used for the representation of the geometry of this scene. We tested both approaches on three different scenes which were obtained by voxelization into two different resolutions of polygonal surface models represented originally as the mesh of triangles in Wavefront Object file format. Results of the tests show that spatio-temporal sparse voxel octree is better solution, when we obtained from 1.73 to 2.08 more compact binary representation, in comparison to the solution when time frames were encoded separately as the set of sparse voxel octrees.

A Retrospective Hydrological Uncertainty Analysis Using Precipitation Estimation Ensembles for a Poorly Gauged Basin in High Mountain Asia

Abstract We study the impact of uncertain precipitation estimates on simulated streamflows for the poorly gauged Yarlung Tsangpo basin (YTB), high mountain Asia (HMA). A process-based hydrological model at 0.5-km resolution is driven by an ensemble of precipitation estimation products (PEPs), including analyzed ground observations, high-resolution precipitation estimates, climate data records, and reanalyses over the 2008–15 control period. The model is then forced retrospectively from 1983 onward to obtain seamless discharge estimates till 2007, a period for which there is very sparse flow data coverage. Whereas temperature forcing is considered deterministic, precipitation is sampled from the predictive distribution, which is obtained through processing PEPs by means of a probabilistic processor of uncertainty. The employed Bayesian processor combines the PEPs and outputs the predictive densities of daily precipitation depth accumulation as well as the probability of precipitation occurrence, from which random precipitation fields for probabilistic model forcing are sampled. The predictive density of precipitation is conditional on the precipitation estimation predictors that are bias corrected and variance adjusted. For the selected HMA study site, discharges simulated from reanalysis and climate data records score lowest against observations at three flow gauging points, whereas high-resolution satellite estimates perform better, but are still outperformed by precipitation fields obtained from analyzed observed precipitation and merged products, which were corrected against ground observations. The applied methodology indicates how missing flows for poorly gauged sites can be retrieved and is further extendable to hydrological projections of climate. Significance Statement We show how to use different precipitation estimates, like computer simulations of weather or satellite observations, in conjunction with all available ground measurements in regions with generally poor meteorological and flow measurement infrastructure. We demonstrate how it is possible to retrieve past unobserved river flows using these estimates in combination with a hydrological computer model for streamflow simulations. The method can help us to better understand the hydrology of poorly gauged regions that play an important role in the distribution of water resources and can be affected by future changes. We applied the method to a large transboundary river basin in China. This basin holds water needed by large, densely populated regions of India that may become water constrained by warmer climate.

GMINN: Gate‐enhanced multi‐space interaction neural networks for click‐through rate prediction

AbstractClick‐through rate (CTR) prediction is a pivotal challenge in recommendation systems. Existing models are prone to disturbances from noise and redundant features, hindering their ability to fully capture implicit and higher‐order feature interactions present in sparse feature data. Moreover, conventional dual‐tower models overlook the significance of layer‐level feature interactions. To address these limitations, this article introduces Gate‐enhanced Multi‐space Interactive Neural Networks (GMINN), a novel model for CTR prediction. GMINN adopts a dual‐tower architecture in which a multi‐space interaction layer is introduced after each layer in the dual‐tower deep neural network. This layer allocates features into multiple subspaces and employs matrix multiplication to establish layer‐level interactions between the dual towers. Simultaneously, a field‐aware gate mechanism is proposed to extract crucial latent information from the original features. Experimental validation on publicly available datasets, Criteo and Avazu, demonstrates the superiority of the proposed GMINN model. Comparative analyses against baseline models reveal that GMINN substantially improves up to 4.09% in AUC and a maximum reduction of 7.21% in Logloss. Additionally, ablation experiments provide further validation of the effectiveness of GMINN.

A Bayesian network for simultaneous keyframe and landmark detection in ultrasonic cine

Accurate landmark detection in medical imaging is essential for quantifying various anatomical structures and assisting in diagnosis and treatment planning. In ultrasound cine, landmark detection is often associated with identifying keyframes, which represent the occurrence of specific events, such as measuring target dimensions at specific temporal phases. Existing methods predominantly treat landmark and keyframe detection as separate tasks without harnessing their underlying correlations. Additionally, owing to the intrinsic characteristics of ultrasound imaging, both tasks are constrained by inter-observer variability, leading to potentially higher levels of uncertainty. In this paper, we propose a Bayesian network to achieve simultaneous keyframe and landmark detection in ultrasonic cine, especially under highly sparse training data conditions. We follow a coarse-to-fine landmark detection architecture and propose an adaptive Bayesian hypergraph for coordinate refinement on the results of heatmap-based regression. In addition, we propose Order Loss for training bi-directional Gated Recurrent Unit to identify keyframes based on the relative likelihoods within the sequence. Furthermore, to exploit the underlying correlation between the two tasks, we use a shared encoder to extract features for both tasks and enhance the detection accuracy through the interaction of temporal and motion information. Experiments on two in-house datasets (multi-view transesophageal and transthoracic echocardiography) and one public dataset (transthoracic echocardiography) demonstrate that our method outperforms state-of-the-art approaches. The mean absolute errors for dimension measurements of the left atrial appendage, aortic annulus, and left ventricle are 2.40 mm, 0.83 mm, and 1.63 mm, respectively. The source code is available at github.com/warmestwind/ABHG.

IterPrompt: An iterative prompt-tuning method for nested relation extraction with dynamic assignment strategy

Most existing relation extraction methods focus on extracting relations with the flat or overlapping structure, where constituent elements of these relations are only entities. However, in many application scenarios, relations with nested structures are more common, of which the elements can be both entities or relation triples. The almost vacant research on nested relation extraction can be attributed to two difficulties: (1) how to formalize and recover the nested structures of relations from the unstructured texts. (2) how to keep the extraction method effective in low-data settings. To solve these problems, first, we formally define the nested RE task and propose a prompt-tuning-based method, called iterPrompt, through constructing the iterative prompt templates for all possible relation triples to establish the nested structures. Then, we combine the prompts with the pre-trained language model by converting the nested RE task into the masked word prediction task to adapt our method to low-data settings. Furthermore, to alleviate the data sparsity problem, a dynamic relation identifier assignment strategy is proposed by associating the candidate relation triple with the randomly generated relation identifier, for representing the semantics of the current extracted relation. Extensive experiments demonstrate that the proposed iterPrompt method outperforms previous state-of-the-art baselines and has good compatibility for both flat and overlapping relation extraction tasks.