Neural Computing and Applications最新文献

Saliency maps are input-resolution matrices used for visualizing local interpretations of image classifiers. Their pixel values reflect the importance of corresponding image locations for the model’s decision. Despite numerous proposals on how to obtain such maps, their evaluation remains an open question. This paper presents a carefully designed experimental procedure along with a set of quantitative interpretation evaluation metrics that rely solely on the original model behavior. Previously noticed evaluation biases have been attenuated by separating locations with high and low values, considering the full saliency map resolution, and using classifiers with diverse accuracies and all the classes in the dataset. We used the proposed evaluation metrics to compare and analyze seven well-known interpretation methods. Our experiments confirm the importance of object background as well as negative saliency map pixels, and we show that the scale of their impact on the model is comparable to that of positive ones. We also demonstrate that a good class score interpretation does not necessarily imply a good probability interpretation. DeepLIFT and LRP-(epsilon) methods proved most successful altogether, while Grad-CAM and Ablation-CAM performed very poorly, even in the detection of positive relevance. The retention of positive values alone in the latter two methods was responsible for the inaccurate detection of irrelevant locations as well.

The identification and region extraction of brain tumors is an essential aspect of clinical image analysis and the diagnosis of brain-related illnesses. The precise and accurate identification of tumors from MRI images is particularly significant in the effective formulating of treatments such as surgery, radiation therapy, and drug therapy. The challenge of segmentation stems from the variability in the size, location, and appearance of tumors, making it a complex task. Various segmentation and classification techniques have been created and designed for brain tumor diagnosis; however, these traditional techniques are time-consuming and subjective and require expertise in image processing. In recent times, deep learning-based approaches have shown promising results in brain tumor segmentation. This research aims to develop a brain tumor segmentation and classification model that enables medical professionals to locate and measure tumors accurately and develop effective treatment and rehabilitation strategies. The process involves segmenting the tumor and further classifying it into its two major types. The parameter estimation from the segmented output provides an insight that is pivotal in the evaluation of MRI brain tumors. With further research and development, deep learning-based segmentation and classification could become an important tool for accurate detection and evaluation of brain tumors. The development of deep learning-based segmentation and classification methods can greatly benefit the medical community, and according to the finding from the experiment, it is shown that the proposed framework excels in brain tumor segmentation and classification with an accuracy of 99.3%.

A primary strategy for constructing click-through rate models based on deep learning involves combining a multi-layer perceptron (MLP) with custom networks that can effectively capture the interactions between different features. This is due to the widespread recognition that relying solely on a vanilla MLP network is not effective in acquiring knowledge about multiplicative feature interactions. These custom networks often employ product methods, such as inner, Hadamard, and outer products, to construct dedicated architectures for this purpose. Among these methods, the outer product has shown superiority in capturing feature interactions. However, the resulting quadratic form from the outer product operation limits the conveyance of informative higher-order interactions to the MLP. Efforts to address this limitation have led to models attempting to increase interaction degrees to higher orders. However, utilizing matrix factorization techniques to reduce learning parameters has resulted in information loss and decreased performance. Furthermore, previous studies have constrained the MLP’s potential by providing it with inputs consisting of homogeneous outer products, thus limiting available information diversity. To overcome these challenges, we introduce DONN, a model that leverages a composite-wise bilinear module incorporating factorized bilinear pooling to mitigate information loss and facilitate higher-order interaction development. Additionally, DONN utilizes a feature-wise bilinear module for outer product computations between feature pairs, augmenting the MLP with combined information. By employing heterogeneous outer products, DONN enhances the MLP’s prediction capabilities, enabling the recognition of additional nonlinear interdependencies. Our evaluation on two benchmark datasets demonstrates that DONN surpasses state-of-the-art models in terms of performance.

Multimodal reasoning is a critical component in the pursuit of artificial intelligence systems that exhibit human-like intelligence, especially when tackling complex tasks. While the chain-of-thought (CoT) technique has gained considerable attention, the existing ScienceQA dataset, primarily focused on multimodal scientific questions and explanations from elementary and high school textbooks, exhibits limitations in providing a comprehensive evaluation across a broader spectrum of open-domain questions. To address this gap, we introduce the COCO Multi-Modal Reasoning (COCO-MMR) dataset, a comprehensive collection of open-ended questions, rationales, and answers derived from the COCO dataset. Unlike previous datasets that rely on multiple-choice questions, our dataset utilizes open-ended questions to more effectively challenge and assess CoT models’ reasoning capabilities. Through comprehensive evaluations and detailed analyses, we demonstrate that our multihop cross-modal attention and sentence-level contrastive learning modules, designed to simulate human thought processes, significantly enhance model comprehension abilities. Experiments confirm the proposed dataset and techniques, showing their potential to advance multimodal reasoning. The data and code are available at https://github.com/weijingxuan/COCO-MMR.

Accurately predicting crop yield is essential for optimizing agricultural practices and ensuring food security. However, existing approaches often struggle to capture the complex interactions between various environmental factors and crop growth, leading to suboptimal predictions. Consequently, identifying the most important feature is vital when leveraging Support Vector Regressor (SVR) for crop yield prediction. In addition, the manual tuning of SVR hyperparameters may not always offer high accuracy. In this paper, we introduce a novel framework for predicting crop yields that address these challenges. Our framework integrates a new hybrid feature selection approach with an optimized SVR model to enhance prediction accuracy efficiently. The proposed framework comprises three phases: preprocessing, hybrid feature selection, and prediction phases. In preprocessing phase, data normalization is conducted, followed by an application of K-means clustering in conjunction with the correlation-based filter (CFS) to generate a reduced dataset. Subsequently, in the hybrid feature selection phase, a novel hybrid FMIG-RFE feature selection approach is proposed. Finally, the prediction phase introduces an improved variant of Crayfish Optimization Algorithm (COA), named ICOA, which is utilized to optimize the hyperparameters of SVR model thereby achieving superior prediction accuracy along with the novel hybrid feature selection approach. Several experiments are conducted to assess and evaluate the performance of the proposed framework. The results demonstrated the superior performance of the proposed framework over state-of-art approaches. Furthermore, experimental findings regarding the ICOA optimization algorithm affirm its efficacy in optimizing the hyperparameters of SVR model, thereby enhancing both prediction accuracy and computational efficiency, surpassing existing algorithms.

The introduction of hybrid alternating current (AC)/direct current (DC) distribution networks led to several developments in smart grid and decentralized power system technology. The paper concentrates on several topics related to the operation of hybrid AC/DC networks. Such as optimization methods, control strategies, energy management, protection issues, and proposed solutions. The implementation of neural network optimization methods has great importance for the successful integration of multiple energy sources, dynamic energy management, establishment of system stability and reliability, power distribution optimization, management of energy storage, and online fault detection and diagnosis in hybrid networks like the hybrid AC–DC microgrids (MG). Taking advantage of renewable energy generation and cost-cutting through the neural network optimization technique holds the key to these progressions. Besides identifying the challenges in the operation of a hybrid system, the paper also compares this system to conventional MGs and shows the benefits of this type of system over different MG structures. This review compares the different topologies, particularly looking at the AC–DC coupled hybrid MGs, and shows the important role of the interlinking of converters that are used for efficient transmission between AC and DC MGs and generally used to implement the different control and optimization techniques. Overall, this review paper can be regarded as a reference, pointing out the pros and cons of integrating hybrid AC/DC distribution networks for future study and improvement paths in this developing area.

A regression model based on graph neural network, tailored for electric circuit dynamics prediction is introduced, providing converter performance predictions on converter circuit level and internal parameter variations. Regardless of the number of components or connections present in a converter circuit, the proposed model can be readily scaled to incorporate different converter circuit topologies. Moreover, the model can be used to analyse converter circuits with any number of circuit components and any control parameters variation. To enable the use of machine learning methods and applications, all physical and switching circuit properties such as converter circuits operating in continuous conduction mode or discontinuous conduction mode are accurately mapped to graph representation. Three of the most common converters (Buck, Boost, and Buck-boost) are used as example circuits applied to model and the target is to predict the gain and current ripples in inductor. The model achieves 99.51% on the (R^2) measure and a mean square error of 0.0263.

Brain metastases present a formidable challenge in cancer management due to the infiltration of malignant cells from distant sites into the brain. Precise segmentation of brain metastases (BM) in medical imaging is vital for treatment planning and assessment. Leveraging deep learning techniques has shown promise in automating BM identification, facilitating faster and more accurate detection. This paper aims to develop an innovative novel deep learning model tailored for BM segmentation, addressing current approach limitations. Utilizing a comprehensive dataset of annotated magnetic resonance imaging (MRI) from Stanford University, the proposed model will undergo thorough evaluation using standard performance metrics. Comparative analysis with existing segmentation methods will highlight the superior performance and efficacy of our model. The anticipated outcome of this research is a highly accurate and efficient deep learning model for brain metastasis segmentation. Such a model holds potential to enhance treatment planning, monitoring, and ultimately improve patient care and clinical outcomes in managing brain metastases.

Data containing noises, outliers, and imbalanced class distributions pose challenges to the traditional classifiers. By incorporating both the membership and non-membership functions, the intuitionistic fuzzy (IF) set has shown potential in designing robust learning algorithms for classifiers. However, the non-membership function used in these IF-based classifiers usually only utilizes the local distribution information of the training samples, and the classifiers are built upon single-hidden layer networks, which degrade the performance of the corresponding classifiers. Broad learning system (BLS) is an emerging neural network model with fast learning speed and flexible network architecture; however, it still fails to distinguish n samples. To this end, in this paper, we propose a new definition of the non-membership function within intuitionistic fuzzy sets and subsequently propose an intuitionistic fuzzy broad learning system (IFBLS) model. The proposed non-membership function incorporates two ratio numbers based on four distances, allowing for the utilization of global information on the distribution of samples and mitigating misclassification of valid samples as noise which is often observed in traditional methods. By using a score function that considers both the membership and non-membership functions to redistribute the importance of the training samples, the proposed IFBLS benefits from both the powerful representation capability of the original BLS and the robust learning of IF-based models. Extensive experiments conducted on 21 imbalanced binary classification problems sourced from the UCI and KEEL repositories illustrate that the proposed IFBLS achieves state-of-the-art performance by attaining the highest testing accuracy in 17 out of the 21 problems.