3,426 research outputs found
Predicting contract participation in the Mekong Delta, Vietnam: A comparison between the artificial neural network and the multinomial logit model
No full textThe research aims of this study are bi-fold: to study factors influencing the uptake of contract farming (CF) and to compare the predicting power of the artificial neural network model (ANN) and the Multinomial Logit Model (MNL) on predicting CF participation in the Mekong Delta, Vietnam. ANN and MNL were employed to analyze on the basis of the transaction cost theory. To validate the ANN, a 10-fold cross-validation procedure was applied to avoid model overfitting. The sensitivity analysis of ANN was used to elicit the magnitude of the correlation between predictors. Multicollinearity was examined with all VIFs lower than two. Among predictors, the most influential roles of the cooperatives and the extension agents/services in supporting CF participation are reported. Also, farmers who conduct frequent access to the market incline to participate in CF. Risk perceptions and preferences are dissimilar across domains, which are also mainly interpreted that risk-averse farmers tend to opt for CF as an effective solution to risks perceived. Thus, heterogeneous approaches should be tailored to promote CF. The findings suggest that MNL outperforms ANN in terms of accuracy percentage and mean absolute error (MAE). However, this result should not be generalized base on the constraint of the data threshold as articulated in the study. The sensitivity analysis of ANN and the estimation results of the MNL relatively agreed on the importance of model predictors. This study is the first to investigate the impacts of the domain-specific risk perceptions and attitudes on CF and also contribute to the debate over the performance between the conventional econometric models versus machine learning techniques.No Full Tex
On-device diagnostic recommendation with heterogeneous federated BlockNets
No full textThe evolution of edge computing has advanced the accessibility of E-health recommendation services, encompassing areas such as medical consultations, prescription guidance, and diagnostic assessments. Traditional methodologies predominantly utilize centralized recommendations, relying on servers to store client data and dispatch advice to users. However, these conventional approaches raise significant concerns regarding data privacy and often result in computational inefficiencies. E-health recommendation services, distinct from other recommendation domains, demand not only precise and swift analyses but also a stringent adherence to privacy safeguards, given the users’ reluctance to disclose their identities or health information. In response to these challenges, we explore a new paradigm called on-device recommendation tailored to E-health diagnostics, where diagnostic support (such as biomedical image diagnostics), is computed at the client level. We leverage the advances of federated learning to deploy deep learning models capable of delivering expert-level diagnostic suggestions on clients. However, existing federated learning frameworks often deploy a singular model across all edge devices, overlooking their heterogeneous computational capabilities. In this work, we propose an adaptive federated learning framework utilizing BlockNets, a modular design rooted in the layers of deep neural networks, for diagnostic recommendation across heterogeneous devices. Our framework offers the flexibility for users to adjust local model configurations according to their device’s computational power. To further handle the capacity skewness of edge devices, we develop a data-free knowledge distillation mechanism to ensure synchronized parameters of local models with the global model, enhancing the overall accuracy. Through comprehensive experiments across five real-world datasets, against six baseline models, within six experimental setups, and various data distribution scenarios, our architecture demonstrates unparalleled performance and robustness in terms of both accuracy and efficiency.Full Tex
Five Years Follow-Up Outcomes of Femtosecond Laser-Assisted Cataract Surgery on Patients with Preexisting Corneal Astigmatism
No full textThi Minh Khanh Pham,1 Xuan Hiep Nguyen,2 Thi Thu Thuy Pham,1 Tran Thanh Hoang3 1Department of Ophthalmology, Hanoi Medical University, Hanoi, Vietnam; 2Eye Center, Tam Anh Hospital, Hanoi, Vietnam; 3Hospital Director, Hadong Eye Hospital, Hanoi, VietnamCorrespondence: Thi Minh Khanh Pham, Department of Ophthalmology, Hanoi Medical University, Number 1 Ton That Tung, Dong Da District, Hanoi, Vietnam, Tel +84-969085588, Email [email protected]: Evaluating the long-term clinical efficacy and safety of femtosecond laser-assisted cataract surgery for correcting corneal astigmatism.Patients and Methods: In this cohort study on follow-up records from preoperative, postoperative 1 week, 1 month, 3 months, 1 year, 3 years, and 5 years, thirty-four eyes with cataract and corneal astigmatism (> 0.50D) were treated with corneal arcuate incisions and femtosecond-laser assisted cataract surgery in Vietnam National Eye Hospital, from January 2017 to February 2023.Results: The rate of postoperative refraction spherical equivalent was within ± 0.50D and ± 1.0D at 3 months (in 91.2% and 100% of the eyes, respectively). The average of preoperative corneal astigmatism was 1.63 ± 0.886D, decreased to 0.53 ± 0.628D in the third month after surgery and stable to 5 years. Surgically induced astigmatism was 1.09 ± 0.413D, which indicated under-correction. However, no complications were recorded.Conclusion: The femtosecond laser-assisted cataract surgery is safe and long effective in correcting the corneal astigmatism in patients with preexisting corneal astigmatism.Keywords: femtosecond laser, corneal astigmatism, cataract, arcuate incisio
Example-based explanations for streaming fraud detection on graphs
No full textFraud detection is one of the most important tasks in Web platforms such as e-commerce, social media, network security, and financial systems. To prevent fraudulent actions from misleading customers or causing significant losses for businesses, various fraud detection methods have been proposed in recent years. However, research on fraud definitions, characteristics, and behaviours has been limited to which users, items, and transactions are considered fraudulent rather than why these entities have been classified as such. This inhibits effective validation of the detected frauds as well as countermeasure design. In this paper, we argue that explanations for discovered frauds may be provided in terms of prior identified frauds. A large variety of comparable frauds would assist investigators to generalise, allowing them to grasp the characteristics that are significant for fraud detection. Feature-annotated graphs are frequently used to detect the type of fraud in which fraudsters commonly interact with a large number of benign users to conceal themselves. Given a fraud subgraph, we propose a query-by-example approach for indexing and extracting the k most similar and diverse fraud subgraphs from prior frauds. To achieve an efficient and adaptive realisation of the approach in a streaming setting, we present a novel graph representation learning technique and discuss the implementation considerations. Comparing our study against baseline techniques revealed that our approach outperforms them in delivering meaningful explanations for various fraud camouflage behaviours.Full Tex
Monitoring agriculture areas with satellite images and deep learning
No full textAgriculture applications rely on accurate land monitoring, especially paddy areas, for timely food security control and support actions. However, traditional monitoring requires field works or surveys performed by experts, which is costly, slow, and sparse. Agriculture monitoring systems are looking for sustainable land use monitoring solutions, starting with remote sensing on satellite data for cheap and timely paddy mapping. The aim of this study is to develop an autonomous and intelligent system built on top of imagery data streams, which is available from low-Earth orbiting satellites, to differentiate crop areas from non-crop areas. However, such agriculture mapping framework poses unique challenges for satellite image processing, including the seasonal nature of crop, the complexity of spectral channels, and adversarial conditions such as cloud and solar radiance. In this paper, we propose a novel multi-temporal high-spatial resolution classification method with an advanced spatio-temporal–spectral deep neural network to locate paddy fields at the pixel level for a whole year long and for each temporal instance. Our method is built and tested on the case study of Landsat 8 data due to its high spatial resolution. Empirical evaluations on real imagery datasets of different landscapes from 2016 to 2018 show the superior of our mapping model against the baselines with over 0.93 F1-score, the importance of each model design, the robustness against seasonal effects, and the visual mapping results.No Full Tex
Isomorphisms in co-TT graphs
No full text2019 Spring.Includes bibliographical references.A threshold tolerance graph is a graph where each vertex v is assigned a weight wv and a tolerance tv, and there is an edge between two vertices vx and vy if and only if wx + wy ≥ min(tx,ty). A co-TT graph is the complement of a threshold tolerance graph. Recognition of these graphs can be done in O(n2) time; however no polynomial-time algorithm to identify isomorphisms between pairs of TT or co-TT graphs was previously known. We give an algorithm to identify these isomorphisms, which takes O(n2) time
Limbal and Conjunctival Epithelial Thickness in Ocular Graft-Versus-Host Disease
No full textPurpose: To compare the thickness of the limbal epithelium (LE) and the bulbar conjunctival epithelium (BCE) between patients with dry eye disease (DED) with and without ocular graft-versus-host disease (GVHD).Methods: This cross-sectional study enrolled 40 patients with moderate to severe DED including 20 with and 20 without chronic ocular GVHD. All patients had a comprehensive clinical ophthalmic assessment. Moreover, the thickness of the LE and BCE in both nasal and temporal regions of both eyes was measured using spectral domain optical coherence tomography.Results: The average LE thickness in all patients with dry eye (GVHD and non-GVHD) was 65.8 +/- 11.9 mu m temporally and 69.7 +/- 11.1 mu m nasally (P = 0.02). The average BCE thickness was 55.8 +/- 11.4 mu m temporally and 60.1 +/- 11.0 mu m nasally (P = 0.03). There were no statistically significant differences between GVHD and non-GVHD groups in LE thickness (69.6 +/- 11.7 vs. 66.1 +/- 6.2 mu m, respectively, P = 0.31) or BCE thickness (58.9 +/- 9.6 vs. 57.3 +/- 9.8 mu m, respectively, P = 0.82). There was a significant correlation between LE thickness and BCE thickness (P = 0.01, Rs = 0.41). A statistically significant negative correlation was also observed between LE thickness and age (P = 0.002, Rs = -0.35). There were no significant correlations between the thickness of the LE or BCE and other clinical parameters.Conclusions: No difference exists in the thickness of the ocular surface epithelia between dry eyes with and without ocular GVHD, which would suggest that these epithelial changes may be independent of the underlying etiology and possibly only reflect the disease severity. Furthermore, there are regional variations in the thickness of the ocular surface epithelia in patients with DED
tt*-geometry and pluriharmonic maps
No full textInternational audienceIn this paper we use the real differential geometric definition of a metric (an unimodular oriented metric) tt*-bundle of Cortés and the author to define a map from the space of metric (unimodular oriented metric) tt*-bundles of rank r over a complex manifold M to the space of pluriharmonic maps from M to (respectively ), where (p,q) is the signature of the metric. In the sequel the image of the map is characterized. It follows, that in signature (r,0) the image of is the whole space of pluriharmonic maps. This generalizes a result of Dubrovin
Resuscitation from Respiratory Arrest Due to Life-Threatening Ventricular Arrhythmias in a Patient with Amitriptyline Intoxication: An Old Problem in a New Era
No full textTan Thanh Nguyen,1 Lac Duy Le,1 Thanh Tri Vu,2 Anh Thai Nguyen,1 Duc Binh Doan,1 Duyen Thi Pham,1 Tung Thanh Pham,1 Chuc Ngoc Vu,3 Minh Hoang Nguyen Vo4 1Cardiac Intensive Department, Thu Duc City Hospital, Ho Chi Minh City, Vietnam; 2Board of Directors, Thu Duc City Hospital, Ho Chi Minh City, Vietnam; 3Emergency Department, Thu Duc City Hospital, Ho Chi Minh City, Vietnam; 4Office of Science Management and International Affairs, Thu Duc City Hospital, Ho Chi Minh City, VietnamCorrespondence: Minh Hoang Nguyen Vo, Office of Science Management and International Affairs, Thu Duc City Hospital, No. 29 Phu Chau Street, Tam Phu ward, Thu Duc, Ho Chi Minh City, 700000, Vietnam, Tel +84 389135014, Email [email protected]: Tricyclic antidepressants (TCAs) were once commonly used to treat major depressive disorder (MDD), but are now considered second-line options after SSRIs and SNRIs. Additionally, TCAs are used to treat other conditions such as chronic pain and enuresis in children. Due to their numerous side effects and potential for drug interactions, cases of poisoning and death from TCA overdose, particularly amitriptyline, are on the rise. Therefore, this article revisits the overview and describes the clinical progression regarding blood gases, ECG, and electrolytes of the patient, as well as the use of 4.2% sodium bicarbonate and 2% lidocaine to treat cases of amitriptyline overdose poisoning.Case Presentation: A 49-year-old female patient was admitted to the hospital due to cardiac and respiratory arrest. The patient had a past medical history of untreated cervical cancer and sleep disorders. Prior to admission, the patient had taken about 20 tablets of amitriptyline 25mg and was in a drowsy state with gasping breaths. During transportation to the hospital, the patient experienced cardiac arrest once and was successfully resuscitated, with a total arrest and resuscitation time of approximately 10 minutes.Results: The use of 4.2% Sodium Bicarbonate and 2% Lidocaine, the patient was not used plasma exchange in this case, proved effective in this case. Continuous monitoring of blood gas levels, ECG, and electrolytes was maintained. The patient was able to walk independently and was discharged after 12 days of treatment.Conclusion: The key factor was the healthcare staff’s quick recognition and timely management of TCA poisoning, in this case, amitriptyline.Keywords: tricyclic antidepressants, amitriptyline, cardiac arrhythmi
performance of the low-rank TT-SVD for large dense tensors on modern multicore CPUs
No full textThere are several factorizations of multidimensional tensors into lower-dimensional components, known as ``tensor networks."" We consider the popular ``tensor-train"" (TT) format and ask, How efficiently can we compute a low-rank approximation from a full tensor on current multicore CPUs? Compared to sparse and dense linear algebra, kernel libraries for multilinear algebra are rare and typically not as well optimized. Linear algebra libraries like BLAS and LAPACK may provide the required operations in principle but often at the cost of additional data movements for rearranging memory layouts. Furthermore, these libraries are typically optimized for the compute-bound case (e.g., square matrix operations), whereas low-rank tensor decompositions lead to memory bandwidth limited operations. We propose a ``TT singular value decomposition"" (TT-SVD) algorithm based on two building blocks: a ``Q-less tall-skinny QR"" factorization and a fused tall-skinny matrix-matrix multiplication and reshape operation. We analyze the performance of the resulting TT-SVD algorithm using the roofline performance model. In addition, we present performance results for different algorithmic variants for shared-memory as well as distributed-memory architectures. Our experiments show that commonly used TT-SVD implementations suffer severe performance penalties. We conclude that a dedicated library for tensor factorization kernels would benefit the community: Computing a low-rank approximation can be as cheap as reading the data twice from main memory. As a consequence, an implementation that achieves realistic performance will move the limit at which one has to resort to randomized methods that only process part of the data.Numerical Analysi
- …
