Applied Intelligence最新文献

Automatically predicting ICD-10 codes from clinical notes using machine learning models can reduce the burden of manual coding. However, existing methods often overlook the semantic relationships between ICD-10 codes, resulting in inaccurate evaluations when clinically similar codes are considered completely different. Traditional evaluation metrics, which rely on equality-based matching, fail to capture the clinical relevance of predicted codes. This study introduces NNBSVR (Neural Network-Based Semantic Vector Representations), a novel approach for generating semantic-based vector representations of ICD-10 codes. Unlike traditional approaches that rely on exact code matching, NNBSVR incorporates contextual and hierarchical information to enhance both prediction accuracy and evaluation methods. We validate NNBSVR using intrinsic and extrinsic evaluation methods. Intrinsic evaluation assesses the vectors’ ability to reconstruct the ICD-10 hierarchy and identify clinically meaningful clusters. Extrinsic evaluation compares our relevancy-based approach, which includes customized evaluation metrics, to traditional equality-based metrics on an ICD-10 code prediction task using a 9.57 million clinical notes corpus. NNBSVR demonstrates significant improvements over equality-based metrics, achieving a 9.81% gain in micro-F1 score on the training set and a 12.73% gain on the test set. A manual review by medical experts on a sample of 10,000 predictions confirms an accuracy of 92.58%, further validating our approach. This study makes two significant contributions: first, the development of semantic-based vector representations that encapsulate ICD-10 code relationships and context; second, the customization of evaluation metrics to incorporate clinical relevance. By addressing the limitations of traditional equality-based evaluation metrics, NNBSVR enhances the automated assignment of ICD-10 codes in clinical settings, demonstrating superior performance over existing methods.

The Graph Convolutional Neural Network (GCN) is a powerful technique for learning and representing graph data, commonly utilized in model-based collaborative filtering recommendation algorithms. However, despite its effectiveness, the issues are data sparsity and interpretability. Most existing GCN-based models simply update the central node’s features by aggregating the features of its neighbors, typically via a weighted sum. Unfortunately, this approach fails to capture the cooperative information hidden in the neighbor interactions. To address this limitation, we propose a recommendation algorithm based on a convolution network of lightweight neighborhood interactive graphs, named the Lightweight Bilinear Graph Convolutional Network (LBGCN). Our approach employs a lightweight graph convolutional neural network as a multi-level feature aggregator, leveraging higher-order connectivity to aggregate neighborhood information into a multi-level feature of the node through the aggregator. Meanwhile, we introduce a local feature aggregator to capture the collaborative filtering signals in the interaction features of neighbors. Finally, we combine the results using an attention mechanism to obtain the embedded representation of final users and items. In addition, we demonstrate the rationality and effectiveness of our proposed model through experiments on three public datasets. The results show that our method could gain 2.52% NDCG improvement at most.

With the widespread application of deep learning, general object detectors have become increasingly popular in our daily lives. Extensive research, however, has shown that existing detectors are vulnerable to patch-based adversarial attacks, which fool such detectors by crafting adversarial patches. Although existing methods have made significant progress in terms of attack success rate, they still suffer from a highly perceptible problem, making it easy for humans to distinguish these evil examples. To address this issue, in this paper, we propose a novel spatial transform-based end-to-end patch attack method, called IPAttack, to synthesize imperceptible adversarial patches. Our approach estimates a flow field (varvec{f}) to formulate adversarial examples rather than introduce small (L_p)-norm constrained external perturbations. Besides, to improve the imperceptibility and maintain a high attack performance, we propose the Object Detector Class Activation Map (OD-CAM) for objectors to extract the most interesting region, which will be applied to spatial transform to generate the final adversarial examples. Extensive experiments demonstrate that the proposed IPAttack can generate patch-wised adversarial examples with high imperceptibility while achieving the best attack performance compared to existing methods.

Large language models (LLMs) have demonstrated extensive capabilities across various natural language processing (NLP) tasks. Knowledge graphs (KGs) harbor vast amounts of facts, furnishing external knowledge for language models. The structured knowledge extracted from KGs must undergo conversion into sentences to align with the input format required by LLMs. Previous research has commonly utilized methods such as triple conversion and template-based conversion. However, sentences converted using existing methods frequently encounter issues such as semantic incoherence, ambiguity, and unnaturalness, which distort the original intent, and deviate the sentences from the facts. Meanwhile, despite the improvement that knowledge-enhanced pre-training and prompt-tuning methods have achieved in small-scale models, they are difficult to implement for LLMs in the absence of computational resources. The advanced comprehension of LLMs facilitates in-context learning (ICL), thereby enhancing their performance without the need for additional training. In this paper, we propose a knowledge prompts generation method, GenKP, which injects knowledge into LLMs by ICL. Compared to inserting triple-conversion or templated-conversion knowledge without selection, GenKP entails generating knowledge samples using LLMs in conjunction with KGs and makes a trade-off of knowledge samples through weighted verification and BM25 ranking, reducing knowledge noise. Experimental results illustrate that incorporating knowledge prompts enhances the performance of LLMs. Furthermore, LLMs augmented with GenKP exhibit superior improvements compared to the methods utilizing triple and template-based knowledge injection.

Spiking Neural Networks (SNNs) are recognized for their energy efficiency due to spike-based communication. In this regard, the shift towards SNNs is driven by their ability to significantly reduce energy consumption while maintaining the performance of ANNs. Converting Artificial Neural Networks (ANNs) to SNNs is a key research focus, but existing methods often struggle with balancing conversion accuracy and latency, and are typically restricted to ReLU activations. We introduce Precision Spiking (PS) neurons, a novel dynamic spiking neuron model that can precisely fit any activation function by jointly regulating spike timing, reset voltage, and membrane potential threshold. This capability enables exact parameter optimization via iterative methods, achieving low-latency, high-accuracy ANN-to-SNN conversion. Experiments on image classification and natural language processing benchmarks confirm state-of-the-art results, with a maximum conversion loss of 0.55% and up to 0.38% accuracy improvement over the original ANN. To the best of our knowledge, this method offers a significant advancement over existing approaches by achieving high-precision fitting of arbitrary activation functions with low latency and minimal conversion loss, thus considerably expanding the range of feasible ANN-to-SNN conversions.

Causality stands as an essential relation for elucidating the reasoning behind given contents. However, current causality knowledge graphs fall short in effectively illustrating the inner logic in a specific domain, i.e. finance. To generate such a functional knowledge graph, we propose the multi-faceted approach encompassing causality detection module, entity linking module, and causality alignment module to automatically construct FinCaKG-Onto with the guidance of expert financial ontology - FIBO. In this paper, we outline the resources and methodology employed for FinCaKG-Onto construction, present the schema of FinCaKG-Onto, and share the final knowledge graph FinCaKG-Onto. Through various user scenarios, we demonstrate that FinCaKG-Onto not only captures nuanced domain expertise but also explicitly unveils the causal logic for any anchor terms. To facilitate your convenience of future use, a check table is conducted as well to showcase the quality of FinCaKG-Onto. The related resources are available in the webpage<https://www.ai.iee.e.titech.ac.jp/FinCaKG-Onto/>.

Due to the scarcity of training samples in the visual tracking task, almost all existing Convolutional Neural Networks (CNNs) based deep tracking algorithms rely heavily on large auxiliary datasets to train the tracking model offline. However, such offline training has two inevitable disadvantages: (1) the learned generic features may be less discriminative for tracking specific objects; (2) the training process demands huge computational power provided by high-performance graphics processing units (GPUs), which is not always available in many practical applications. Therefore, learning effective generic features without offline training for robust visual tracking is a necessary and challenging task. This paper tackles this task by proposing the Sparse Filters-based Convolutional Network (SFCN), which is a fully feed-forward convolutional network with a lightweight structure including two convolutional layers. Its convolutional kernels are a set of sparse filters learned and updated online from local patches using sparse dictionary learning. Benefiting from the learned sparse filters, SFCN learns effective generic features by exploiting both the discriminative information between the foreground and background of the target region and the hierarchical layout information among the local patches inside each target candidate region. Furthermore, a dynamic model updating strategy is adopted to alleviate the drift problem. Extensive experiments on five large-scale benchmark datasets show that the proposed method performs favorably against several state-of-the-art tracking algorithms.

To accurately predict the electricity consumption trend of individual users and even the entire industry, this paper studies an intelligent prediction algorithm for the power industry based on time series and entropy weight method. Using ARIMA model and X12 model to establish a monthly electricity consumption prediction model, the study obtains the monthly electricity consumption prediction value for the expansion of the power industry. The entropy weight method is employed to calculate the weights of two power industry expansion month electricity consumption forecasting models, thereby achieving intelligent forecasting. The experimental results demonstrate that the maximum error of the proposed method is only 1.78%, and the average time complexity and average space complexity of the proposed algorithm are both below the set threshold.

Emergency decision-making problem is common in our daily life. To solve this kind of problem, a group of decision-makers (DMs) are usually invited to make a decision in a limited time. Since multiple attributes are usually considered, it’s called large-scale multi-attribute emergency group decision-making (LS-MA-EGDM). There are two issues in the general research of LS-MA-EGDM. First, clustering and consensus-reaching process (CRP) should consider the influence of DMs’ intrinsic features. Second, consensus adjustment within and among sub-clusters ought to be fast to prevent multi-round iteration. Accordingly, (1) we introduce graph attention mechanism to calculate the attention coefficients between DM pair’s intrinsic features. The multi-head graph attention coefficient based on social network analysis (SNA) is proposed, which is then combined with opinion similarity to construct a social network clustering method. (2) The Einstein product operator is introduced to propagate the attention coefficients and yield DMs’ weights, which is then incorporated in the subsequent adjustment allocation. (3) Identification rules are provided based on four consensus types in the CRP. The one-iteration personalized adjustment strategies corresponding to different consensus types are then proposed. (4) Evidential reasoning (ER) algorithm is finally utilized to aggregate the preferences of clusters after consensus is reaching. The proposed method is further applied to a chemical plant explosion in Texas to illustrate its effectiveness and validity in dealing with emergencies.