Optimizing trigger timing in minimal ovarian stimulation for In Vitro fertilization using machine learning models with random search hyperparameter tuning.

Various studies have emphasized the importance of identifying the optimal Trigger Timing (TT) for the trigger shot in In Vitro Fertilization (IVF), which is crucial for the successful maturation and release of oocytes, especially in minimal ovarian stimulation treatments. Despite its significance for the ultimate success of IVF, determining the precise TT remains a complex challenge for physicians due to the involvement of multiple variables. This study aims to enhance TT by developing a machine learning multi-output model that predicts the expected number of retrieved oocytes, mature oocytes (MII), fertilized oocytes (2 PN), and useable blastocysts within a 48-h window after the trigger shot in minimal stimulation cycles. By utilizing this model, physicians can identify patients with possible early, late, or on-time trigger shots. The study found that approximately 27 % of treatments administered the trigger shot on a suboptimal day, but optimizing the TT using the developed Artificial Intelligence (AI) model can potentially increase useable blastocyst production by 46 %. These findings highlight the potential of predictive models as a supplementary tool for optimizing trigger shot timing and improving IVF outcomes, particularly in minimal ovarian stimulation. The experimental results underwent statistical validation, demonstrating the accuracy and performance of the model. Overall, this study emphasizes the value of AI prediction models in enhancing TT and making the IVF process safer and more efficient.

A Multi-Layer Feature Fusion Method for Few-Shot Image Classification.

In image classification, few-shot learning deals with recognizing visual categories from a few tagged examples. The degree of expressiveness of the encoded features in this scenario is a crucial question that needs to be addressed in the models being trained. Recent approaches have achieved encouraging results in improving few-shot models in deep learning, but designing a competitive and simple architecture is challenging, especially considering its requirement in many practical applications. This work proposes an improved few-shot model based on a multi-layer feature fusion (FMLF) method. The presented approach includes extended feature extraction and fusion mechanisms in the Convolutional Neural Network (CNN) backbone, as well as an effective metric to compute the divergences in the end. In order to evaluate the proposed method, a challenging visual classification problem, maize crop insect classification with specific pests and beneficial categories, is addressed, serving both as a test of our model and as a means to propose a novel dataset. Experiments were carried out to compare the results with ResNet50, VGG16, and MobileNetv2, used as feature extraction backbones, and the FMLF method demonstrated higher accuracy with fewer parameters. The proposed FMLF method improved accuracy scores by up to 3.62% in one-shot and 2.82% in five-shot classification tasks compared to a traditional backbone, which uses only global image features.

A novel methodology for optimal location of reactive compensation through deep neural networks.

The present investigation proposes a methodology for the optimal location of reactive compensation in an electrical power system (EPS) through deep neural networks for voltage profile improvement. One of the main parameters to consider regarding EPS reliability is the voltage profile, a parameter that can be affected due to unexpected increases in impedance and loads in the system that translate as overloads in the system and an increase in the number of users. A voltage profile below the minimum or above the maximum accepted in the regulations of each country puts at risk the correct operation of equipment connected to the electrical network and, in turn, can cause economic losses and human lives (e.g by not guaranteeing reliability for hospitals and similar institutions). Economically, one of the most viable alternatives for improving voltage profiles is reactive compensation which in itself is carried out through capacitor banks. Therefore, this work proposes to find the correct location of capacitor banks in an electrical power system (using IEEE 14, 30 and 118 bus-bars systems as cases of study). In each system, the highest reactive load is identified, thus three values for reactive compensation are established as 80%, 50% and 25% of this maximum. Then, with these values, power flows are generated by locating each one of the reactive compensators' possible values in each one of the bars of the system, hence generating a large number of training data so that finally the neural network is capable of providing a quantitative classification highlighting which compensation and in which bus-bar produces the best result. The result is assessed by applying a modified standard deviation which evaluates the separation of the voltage profiles from the ideal desired value of 1pu.

Probing the interfacial strength of novel multi-layer zirconias.

OBJECTIVES: The rapidly increasing use of zirconia-based CAD/CAM multi-layer structures in dentistry calls for a thorough evaluation of their mechanical integrity. This work examines the effect of the multi-layering architecture as well as variations in composition and inclusion of pigments among the layers on the flexural strength of multi-layer zirconias. METHODS: A modified 4-point bending test, aided by a Finite Element Analysis (FEA), was used to probe the interfacial strength of 3 classes of yttria-partially-stabilized zirconia: Ultra Translucent Multi-Layer (UTML-5Y-PSZ), Super Translucent Multi-Layer (STML-4Y-PSZ), Multi-Layer (ML-3Y-PSZ). In accord with the size limitation (22-mm height) of CAD/CAM pucks, test samples were prepared in the form of "long" (25×2×3mm) and "short" (17.8×1.5×2mm) beams. Homogeneous beams (both long and short) were produced from either the Enamel (the lightest shade) or Dentin (the darkest shade) layer, whereas multi-layer beams (short beam only) were obtained by cutting the pucks along their thickness direction, where the material components of various shades were stacked. RESULTS: The Enamel and Dentin layers exhibited similar flexural strength for a given material class, with ML amassing the highest strength (800-900MPa) followed by STML (560-650MPa) and UTML (470-500MPa). The 3 classes of multi-layer zirconia showed a trade-off between strength and translucency, reflecting different yttria contents in these materials. The failure stress of the cross-sectional multi-layer beams was, however, ∼30% lower than that of their Enamel or Dentin layer counterparts, regardless of material tested. SIGNIFICANCE: The weakness of interfaces is a drawback in these materials. Additionally, when measuring strength using short beam flexure, friction between the specimen and supporting pins and accuracy in determining loading span distances may lead to major errors.

Unfolding the Complexity of the Global Value Chain: Strength and Entropy in the Single-Layer, Multiplex, and Multi-Layer International Trade Networks.

The worldwide trade network has been widely studied through different data sets and network representations with a view to better understanding interactions among countries and products. Here we investigate international trade through the lenses of the single-layer, multiplex, and multi-layer networks. We discuss differences among the three network frameworks in terms of their relative advantages in capturing salient topological features of trade. We draw on the World Input-Output Database to build the three networks. We then uncover sources of heterogeneity in the way strength is allocated among countries and transactions by computing the strength distribution and entropy in each network. Additionally, we trace how entropy evolved, and show how the observed peaks can be associated with the onset of the global economic downturn. Findings suggest how more complex representations of trade, such as the multi-layer network, enable us to disambiguate the distinct roles of intra- and cross-industry transactions in driving the evolution of entropy at a more aggregate level. We discuss our results and the implications of our comparative analysis of networks for research on international trade and other empirical domains across the natural and social sciences.

Aplicação de técnicas de redes neurais e modelagem atmosférica para elaboração de previsões de vazão na Bacia do Rio Grande (MG) / Application of neural networks techniques and atmospheric modeling to elaborate inflow forecasts in the Rio Grande Basin (MG), Brazil

RESUMO Neste estudo foi proposta a elaboração de um modelo de previsão de vazões no horizonte de dez dias para a Usina Hidrelétrica de Furnas, localizada na Bacia do Rio Grande, Minas Gerais, a partir da aplicação de redes neurais artificiais (RNA), informações de vazão natural e precipitação observada e prevista. O modelo foi desenvolvido utilizando o software Matlab(r) Neural Network Toolbox. Escolheu-se uma rede neural do tipo perceptron multicamadas (MLP), treinada com algoritmo supervisionado de retropropagação Levenberg-Marquardt. As previsões de precipitação foram obtidas a partir do modelo ETA/Centro de Previsão do Tempo e Estudos Climáticos (CPTEC), e utilizadas com e sem tratamento matemático. Foram realizados três experimentos, dividindo-se o histórico de dados em três períodos, sendo o primeiro para a calibração do modelo, o segundo para a validação e o terceiro para os testes. Em cada experimento foi variado o conjunto de dados de entrada, sendo utilizada, no primeiro experimento, somente a vazão passada para prever os dez dias de vazão futura. No segundo foi adicionada a precipitação observada e, no terceiro, a previsão de precipitação. Os resultados da modelagem chuva-vazão obtidos com a previsão de precipitaçãodo modelo ETA não apresentaram melhorias estatísticas em comparação com os experimentos que só utilizaram informações passadas. No entanto, quando se utilizou a previsão de precipitação corrigida matematicamente, observou-se uma melhora sensível tanto nos índices estatísticos quanto na representação da previsão simulada no hidrograma, ficando o desempenho da modelagem proposta neste estudo semelhante à encontrada em modelos conceituais do tipo chuva-vazão.

ABSTRACT The purpose of this study was to elaborate a ten-year runoff forecast model for the Furnas hydroelectric plant. The facility is located in the Rio Grande Basin in the state of Minas Gerais, Brazil. Artificial neural networks were used to determine natural flow as well as observed and predicted precipitation. The model was created using the Matlab(r) Neural Network Toolbox software, and the multi-layers perceptron (MLP) was trained with supervised learning algorithm Levenberg-Marquardt. Precipitation forecasts derived from ETA/Centro de Previsão do Tempo e Estudos Climáticos (CPTEC) model, and both raw and mathematical adjusted data were used. Historical data was separated in three different periods in order to calibrate, validate and test the model. The first share was used for calibration, the second portion was used for validation and the third one to test the model. In each experiment the input data was modified; thus, in the first experiment, to forecast the ten day runoff, only the past runoff data was considered. In the second experiment, observed precipitation was added; and in the third one, the forecast precipitation was added. The rainfall-runoff modeling results did not show any significant improvement in the statistics when ETA input data is compared with the experiments that only used past information as input. Nevertheless, when forecast precipitation was used with mathematical adjustment, a mild improvement was shown for the statistics index and for the forecast hydrogram simulation. As a result, the modeling performance proposed in this study is similar to that found in conceptual models of rainfall-runoff type.

Funciones de transferencia en el perceptrón multicapa: efecto de su combinación en entrenamiento local y distribuido / Transfer functions in the multilayer perceptron: effects of its combination on local and distributed training

El perceptrón multicapa (PMC) figura dentro de los tipos de redes neuronales artificiales (RNA) con resultados útiles en los estudios de relación estructura-actividad. Dado que los volúmenes de datos en proyectos de Bioinformática son eventualmente grandes, se propuso evaluar algoritmos para acortar el tiempo de entrenamiento de la red sin afectar su eficiencia. Se desarrolló un algoritmo para el entrenamiento local y distribuido del PMC con la posibilidad de variar las funciones de transferencias para lo cual se utilizaron el Weka y la Plataforma de Tareas Distribuidas Tarenal para distribuir el entrenamiento del perceptrón multicapa. Se demostró que en dependencia de la muestra de entrenamiento, la variación de las funciones de transferencia pueden reportar resultados mucho más eficientes que los obtenidos con la clásica función Sigmoidal, con incremento de la g-media entre el 4.5 y el 17 por ciento. Se encontró además que en los entrenamientos distribuidos es posible alcanzar eventualmente mejores resultados que los logrados en ambiente local(AU)

The multilayer perceptron (PMC) ranks among the types of artificial neural networks (ANN), which has provided better results in studies of structure-activity relationship. As the data volumes in Bioinformatics' projects are eventually big, it was proposed to evaluate algorithms to shorten the training time of the network without affecting its efficiency. There were evaluated different tools that work with ANN and were selected Weka algorithm for extracting the network and the Platform for Distributed Task Tarenal to distribute the training of multilayer perceptron. Finally, it was developed a training algorithm for local and distributed the MLP with the possibility of varying transfer functions. It was shown that depending on the training sample, the change of transfer functions can yield results much more efficient than those obtained with the classic sigmoid function with increased g-media between 4.5 and 17 percent. Moreover, it was found that with distributed training can be achieved eventually, better results than those achieved in the local environment(AU)