Cervical cancer, like many other cancers, is most treatable when detected at an early stage. Using classification methods helps find early signs of cancer and small tumors. This allows doctors to act quickly and offer treatments that might cure the cancer. This study presents a comprehensive approach to the classification of squamous cell carcinoma (SCC) leveraging a dataset comprising 1140 single-cell images sourced from Herlev. A variety of deep learning models, including DenseNet121, DenseNet169, InceptionResNet, XceptionNet, ResNet50, and ResNet101, are employed both individually and in ensembles, demonstrating their efficacy in classifying diverse cellular features. To validate the robustness of results, k-fold cross-validation is conducted, further affirming the effectiveness of the proposed methodology. Thorough exploration produces a precise and effective model for SCC classification, providing detailed insights into both normal and abnormal cell types. These findings show that transfer learning-based deep neural networks and ensemble methods can improve the diagnostic capabilities by 98% accuracy, of SCC classification systems for different cell types.

Semantic Role Labeling (SRL) serves as the foundational and pivotal technology for semantic analysis. However, methods relying on pre-trained language models are constrained by issues such as semantic ambiguity and training complexity. Addressing this concern, this paper proposes a Chinese SRL approach that integrates pre-trained language models with Biaffine technology, aiming to enhance the model’s capability in processing semantic information from long sentences while reducing training complexity. By incorporating pooling techniques and part-of-speech features, the model exhibits significant improvements in capturing semantic role boundary relationships. Experimental results demonstrate that the RoBERTa-MPBF model employing maximum pooling achieves an F1 score of 90.89% on the CPB dataset, outperforming models solely based on conditional random fields. Moreover, the introduction of part-of-speech tagging results in an average F1 score improvement of approximately 1.5%. Despite the increased computational burden, considering the performance enhancement, this additional time cost is deemed acceptable. In convolutional kernel size testing, the model maintains F1 scores between 88.6% and 88.8% when the kernel size is 2 or 3. However, as the kernel size increases to 4, the F1 score drops to 80.37%, and further increases to 5 result in an F1 score reduction to 69.51%.

As agricultural machinery evolves towards intelligence and automation, obstacle detection in agricultural environments becomes crucial for safe operations of intelligent agricultural machinery. Pedestrians, as one of the most common obstacles in orchards, usually exhibit autonomy and behavioral unpredictability. Therefore, the development of intelligent agriculture requires reliable pedestrian detection technology. To address this, we propose ReB-DINO, a robust and accurate orchard pedestrian object detection model based on an improved DINO. Initially, we improve the feature extraction module of DINO using structural re-parameterization, enhancing accuracy and speed of the model through training and decoupling inference. In addition, a progressive feature fusion module is employed to fuse the extracted features and improve model accuracy. Finally, the network incorporates a convolutional block attention mechanism and an improved loss function to improve pedestrian detection rates. The experimental results demonstrate a 1.6% improvement in Recall on the NREC dataset compared to the baseline. Moreover, the results show a 4.2% improvement in mAP and the number of parameters decreases by 40.2% compared to the original DINO, enhancing accuracy and real-time object detection in apple orchards while maintaining lightweight attributes, surpassing mainstream object detection models.

A knowledge graph (KG) is a multidirectional labelled graph used for the graphical representation of knowledge where each node represents an entity, and the edge connecting the two nodes represents a relationship. There has been a rise in the popularity of using KG in information retrieval, recommender system, dialogue system, and question-answering system (QAS). The QAS can be either over structured data or unstructured data. This article studies the existing techniques for question-answering systems over KG (KGQA). We collect more than 30 articles on the question-answering system over a knowledge graph. We give a brief introduction to KG and QAS. Further, we discuss challenges, datasets and evaluation metrics used to evaluate techniques. Finally, we conclude the article by discussing some open research aspects, highlighting factors of the low-resource language for KGQA task and remarks on existing systems.

This study analyzed the temporal variation of scour hole dimensions caused by a plain wall jet, which is one of the most hazardous issues faced by hydraulic structures. The study employed two recently developed artificial intelligence-based models, Extreme Learning Machine (ELM) and Multi-Gen Genetic Programming (MGGP), to predict scour hole dimensions and identify effective parameters. Both models accurately predicted the scour hole dimensions, with MGGP outperforming ELM for both training and testing data. MGGP presented four equations that can be used by designers to predict the temporal variations of scour hole dimensions with high accuracy. The non-dimensional form of the scouring time was found to be the most effective parameter, while the channel width ratio and standard deviation of sediments had negligible effects on the accuracy of the models. The study found that the effectiveness of the densimetric Froude number should be considered for predicting the temporal variation of scour hole dimensions due to plain wall jets. The proposed equations from both models had higher accuracy than previous empirical models. Overall, this study provides valuable insights into predicting and mitigating jet scour problems in hydraulic structures.

This study presents a framework for detecting and mitigating fake and potentially attacking user communities within 5G social networks. This framework utilizes geo-location information, community trust within the network, and AI community detection algorithms to identify users that can cause harm. The framework incorporates an artificial control model to select appropriate community detection algorithms and employs a trust-based strategy to identify and filter out potential attackers. It adapts its approach by utilizing user and attack requirement data through the artificial conscience control model while considering the dynamics of community trust within the network. What sets this framework apart from other fake user detection mechanisms is its capacity to consider attributes challenging for malicious users to mimic. These attributes include the trust established within the community over time, the geographical location, and the framework's adaptability to different attack scenarios. To validate its efficacy, we apply the framework to synthetic social network data, demonstrating its ability to distinguish potential malicious users from trustworthy ones.

The quality of machine learning (ML) models deployed in dynamic environments tends to decline over time due to disparities between the data used for training and the upcoming data available for prediction, which is commonly known as drift. Therefore, it is important for ML models to be capable of detecting any changes or drift in the data distribution and updating the ML model accordingly. This study presents various drift detection techniques to identify drift in the survival outcomes of patients who underwent cardiac surgery. Additionally, this study proposes several drift adaptation strategies, such as adaptive learning, incremental learning, and ensemble learning. Through a detailed analysis of the results, the study confirms the superior performance of ensemble model, achieving a minimum mean absolute error (MAE) of 10.684 and 2.827 for predicting hospital stay and ICU stay, respectively. Furthermore, the models that incorporate a drift adaptive framework exhibit superior performance compared to the models that do not include such a framework.

Labeling fine-grained objects manually is extremely challenging, as it is not only label-intensive but also requires professional knowledge. Accordingly, robust learning methods for fine-grained recognition with web images collected from Internet of Things have drawn significant attention. However, training deep fine-grained models directly using untrusted web images is confronted by two primary obstacles: 1) label noise in web images and 2) domain variance between the online sources and test datasets. To this end, in this study, we mainly focus on addressing these two pivotal problems associated with untrusted web images. To be specific, we introduce an end-to-end network that collaboratively addresses these concerns in the process of separating trusted data from untrusted web images. To validate the efficacy of our proposed model, untrusted web images are first collected by utilizing the text category labels found within fine-grained datasets. Subsequently, we employ the designed deep model to eliminate label noise and ameliorate domain mismatch. And the chosen trusted web data are utilized for model training. Comprehensive experiments and ablation studies validate that our method consistently surpasses other state-of-the-art approaches for fine-grained recognition task in a real-world scenario. Simultaneously, this introduces a novel pipeline for fine-grained recognition with substantial efficacy in practical applications. The source code and models can be accessed at: https://github.com/NUST-Machine-Intelligence-Laboratory/DDN.