Comprehensive Experiments Research Articles

EHNet: Efficient Hybrid Network with Dual Attention for Image Deblurring

The motion of an object or camera platform makes the acquired image blurred. This degradation is a major reason to obtain a poor-quality image from an imaging sensor. Therefore, developing an efficient deep-learning-based image processing method to remove the blur artifact is desirable. Deep learning has recently demonstrated significant efficacy in image deblurring, primarily through convolutional neural networks (CNNs) and Transformers. However, the limited receptive fields of CNNs restrict their ability to capture long-range structural dependencies. In contrast, Transformers excel at modeling these dependencies, but they are computationally expensive for high-resolution inputs and lack the appropriate inductive bias. To overcome these challenges, we propose an Efficient Hybrid Network (EHNet) that employs CNN encoders for local feature extraction and Transformer decoders with a dual-attention module to capture spatial and channel-wise dependencies. This synergy facilitates the acquisition of rich contextual information for high-quality image deblurring. Additionally, we introduce the Simple Feature-Embedding Module (SFEM) to replace the pointwise and depthwise convolutions to generate simplified embedding features in the self-attention mechanism. This innovation substantially reduces computational complexity and memory usage while maintaining overall performance. Finally, through comprehensive experiments, our compact model yields promising quantitative and qualitative results for image deblurring on various benchmark datasets.

Semantic-aware frame-event fusion based pattern recognition via large vision–language models

Pattern recognition through the fusion of RGB frames and Event streams has emerged as a novel research area in recent years. Current methods typically employ backbone networks to individually extract the features of RGB frames and event streams, and subsequently fuse these features for pattern recognition. However, we posit that these methods may suffer from two key issues: (1). They attempt to directly learn a mapping from the input vision modality to the semantic labels. This approach often leads to sub-optimal results due to the disparity between the input and semantic labels; (2). They utilize small-scale backbone networks for the extraction of RGB and Event input features, thus these models fail to harness the recent performance advancements of large-scale visual-language models. In this study, we introduce a novel pattern recognition framework that consolidates the semantic labels, RGB frames, and event streams, leveraging pre-trained large-scale vision–language models. Specifically, given the input RGB frames, event streams, and all the predefined semantic labels, we employ a pre-trained large-scale vision model (CLIP vision encoder) to extract the RGB and event features. To handle the semantic labels, we initially convert them into language descriptions through prompt engineering and polish using ChatGPT, and then obtain the semantic features using the pre-trained large-scale language model (CLIP text encoder). Subsequently, we integrate the RGB/Event features and semantic features using multimodal Transformer networks. The resulting frame and event tokens are further amplified using self-attention layers. Concurrently, we propose to enhance the interactions between text tokens and RGB/Event tokens via cross-attention. Finally, we consolidate all three modalities using self-attention and feed-forward layers for recognition. Comprehensive experiments on the HARDVS and PokerEvent datasets fully substantiate the efficacy of our proposed SAFE model. The source code has been released at https://github.com/Event-AHU/SAFE_LargeVLM.

GGAS2SN: Gated Graph and SmilesToSeq Network for Solubility Prediction.

Aqueous solubility is a critical physicochemical property of drug discovery. Solubility is a key issue in pharmaceutical development because it can limit a drug's absorption capacity. Accurate solubility prediction is crucial for pharmacological, environmental, and drug development studies. This research introduces a novel method for solubility prediction by combining gated graph neural networks (GGNNs) and graph attention neural networks (GATs) with Smiles2Seq encoding. Our methodology involves converting chemical compounds into graph structures with nodes representing atoms and edges indicating chemical bonds. These graphs are then processed by using a specialized graph neural network (GNN) architecture. Incorporating attention mechanisms into GNN allows for capturing subtle structural dependencies, fostering improved solubility predictions. Furthermore, we utilized the Smiles2Seq encoding technique to bridge the semantic gap between molecular structures and their textual representations. Smiles2Seq seamlessly converts chemical notations into numeric sequences, facilitating the efficient transfer of information into our model. We demonstrate the efficacy of our approach through comprehensive experiments on benchmark solubility data sets, showcasing superior predictive performance compared to traditional methods. Our model outperforms existing solubility prediction models and provides interpretable insights into the molecular features driving solubility behavior. This research signifies an important advancement in solubility prediction, offering potent tools for drug discovery, formulation development, and environmental assessments. The fusion of GGNN and Smiles2Seq encoding establishes a robust framework for accurately forecasting solubility across various chemical compounds, fostering innovation in various domains reliant on solubility data.

A multi-source domain feature adaptation network for potato disease recognition in field environment.

Accurate identification of potato diseases is crucial for reducing yield losses. To address the issue of low recognition accuracy caused by the mismatch between target domain and source domain due to insufficient samples, the effectiveness of Multi-Source Unsupervised Domain Adaptation (MUDA) method in disease identification is explored. A Multi-Source Domain Feature Adaptation Network (MDFAN) is proposed, employing a two-stage alignment strategy. This method first aligns the distribution of each source-target domain pair within multiple specific feature spaces. In this process, multi-representation extraction and subdomain alignment techniques are utilized to further improve alignment performance. Secondly, classifier outputs are aligned by leveraging decision boundaries within specific domains. Taking into account variations in lighting during image acquisition, a dataset comprising field potato disease images with five distinct disease types is created, followed by comprehensive transfer experiments. In the corresponding transfer tasks, MDFAN achieves an average classification accuracy of 92.11% with two source domains and 93.02% with three source domains, outperforming all other methods. These results not only demonstrate the effectiveness of MUDA but also highlight the robustness of MDFAN to changes in lighting conditions.

YOLO-RWY: A Novel Runway Detection Model for Vision-Based Autonomous Landing of Fixed-Wing Unmanned Aerial Vehicles

In scenarios where global navigation satellite systems (GNSSs) and radio navigation systems are denied, vision-based autonomous landing (VAL) for fixed-wing unmanned aerial vehicles (UAVs) becomes essential. Accurate and real-time runway detection in VAL is vital for providing precise positional and orientational guidance. However, existing research faces significant challenges, including insufficient accuracy, inadequate real-time performance, poor robustness, and high susceptibility to disturbances. To address these challenges, this paper introduces a novel single-stage, anchor-free, and decoupled vision-based runway detection framework, referred to as YOLO-RWY. First, an enhanced data augmentation (EDA) module is incorporated to perform various augmentations, enriching image diversity, and introducing perturbations that improve generalization and safety. Second, a large separable kernel attention (LSKA) module is integrated into the backbone structure to provide a lightweight attention mechanism with a broad receptive field, enhancing feature representation. Third, the neck structure is reorganized as a bidirectional feature pyramid network (BiFPN) module with skip connections and attention allocation, enabling efficient multi-scale and across-stage feature fusion. Finally, the regression loss and task-aligned learning (TAL) assigner are optimized using efficient intersection over union (EIoU) to improve localization evaluation, resulting in faster and more accurate convergence. Comprehensive experiments demonstrate that YOLO-RWY achieves AP50:95 scores of 0.760, 0.611, and 0.413 on synthetic, real nominal, and real edge test sets of the landing approach runway detection (LARD) dataset, respectively. Deployment experiments on an edge device show that YOLO-RWY achieves an inference speed of 154.4 FPS under FP32 quantization with an image size of 640. The results indicate that the proposed YOLO-RWY model possesses strong generalization and real-time capabilities, enabling accurate runway detection in complex and challenging visual environments, and providing support for the onboard VAL systems of fixed-wing UAVs.

GSSCL: A framework for Graph Self-Supervised Curriculum Learning based on clustering label smoothing

Graph self-supervised learning is an effective technique for learning common knowledge from unlabeled graph data through pretext tasks. To capture the interrelationships between nodes and their essential roles globally, existing methods use clustering labels as self-supervised signals. However, in some cases, these methods may introduce noise, leading to over-fitting of the model and a reduction in performance. To address these issues, a novel framework for Graph Self-Supervised Curriculum Learning based on clustering label smoothing called GSSCL has been proposed. GSSCL clusters knowledge in an easy-to-difficult manner, reducing the heavy dependence on the reliability of clustering and improving the generalizability of the model. Moreover, the Silhouette Coefficient is employed to evaluate the clustering confident scores for all nodes. Some nodes are selected based on high confident scores to perform self-supervised learning. To account for the possibility of complex heterophilous information in graphs (e.g., noisy links), clustering pseudo-label smoothing is performed on K-nearest neighbor graphs built upon the similarities between node features instead of the original graph structures. The obtained multi-scale knowledge is then applied to curriculum learning. Finally, comprehensive experiments conducted across diverse public graph benchmarks demonstrate the superior performance of the proposed framework. It exhibits comparable results to state-of-the-art methods across semi-supervised node classification and clustering tasks.

Advancing epigenetic profiling in cervical cancer: machine learning techniques for classifying DNA methylation patterns.

This study investigates the ability to predict DNA methylation patterns in cervical cancer cells using decision-tree-based ensemble approaches and neural network-based models. The research findings suggest that a model based on random forest achieves a significant prediction accuracy of 91.35%. This projection was derived from comprehensive experimentation and a meticulous performance evaluation of the random forest model, employing a range of measures including Accuracy, Sensitivity, Specificity, Matthews Correlation Coefficient, F1-score, Recall, and Precision. The results indicate that the random forest model exhibits superior performance compared to other tree-based models such as the Simple Decision Tree and XGBoost, as well as neural network-based models including Convolutional Neural Networks, Feed Forward Networks, and Wavelet Neural Networks. The findings indicate that using random forest-based techniques has great potential for future study and might be highly valuable in clinical applications, especially in improving diagnostic and treatment strategies based on epigenetic profiles.

Optimizing cancer classification: a hybrid RDO-XGBoost approach for feature selection and predictive insights

The identification of relevant biomarkers from high-dimensional cancer data remains a significant challenge due to the complexity and heterogeneity inherent in various cancer types. Conventional feature selection methods often struggle to effectively navigate the vast solution space while maintaining high predictive accuracy. In response to these challenges, we introduce a novel feature selection approach that integrates Random Drift Optimization (RDO) with XGBoost, specifically designed to enhance the performance of cancer classification tasks. Our proposed framework not only improves classification accuracy but also offers valuable insights into the underlying biological mechanisms driving cancer progression. Through comprehensive experiments conducted on real-world cancer datasets, including Central Nervous System (CNS), Leukemia, Breast, and Ovarian cancers, we demonstrate the efficacy of our method in identifying a smaller subset of unique and relevant genes. This selection results in significantly improved classification efficiency and accuracy. When compared with popular classifiers such as Support Vector Machine, K-Nearest Neighbor, and Naive Bayes, our approach consistently outperforms these models in terms of both accuracy and F-measure metrics. For instance, our framework achieved an accuracy of 97.24% in the CNS dataset, 99.14% in Leukemia, 95.21% in Ovarian, and 87.62% in Breast cancer, showcasing its robustness and effectiveness across different types of cancer data. These results underline the potential of our RDO-XGBoost framework as a promising solution for feature selection in cancer data analysis, offering enhanced predictive performance and valuable biological insights.

Coupling effects of irrigation amount and fertilization rate on growth and bioactive components of four-year-old licorice (Glycyrrhiza uralensis Fisch) in arid regions of Xinjiang

Water scarcity, over-fertilization, and improper crop management practices severely limit the sustainable cultivation of licorice (Glycyrrhiza uralensis Fisch) in the arid regions of Xinjiang. To elucidate the impacts of integrated water and fertilizer management on the growth characteristics and bioactive components (glycyrrhizic acid and liquiritin) of four-year-old licorice plants, a comprehensive four-year field experiment was conducted from 2019 to 2022.The experiment included four irrigation levels (W1: 2500 m³/ha, W2: 4000 m³/ha, W3: 5500 m³/ha, W4: 7000 m³/ha) and four fertilization rates (F1: 305 kg/ha, F2: 610 kg/ha, F3: 915 kg/ha, F4: 1220 kg/ha), following a completely randomized design. Results indicated that both irrigation and fertilization significantly influenced plant height, root length, root weight, root diameter, leaf area index, and root-to-shoot ratio. The optimal growth characteristics were observed under the W2F2 treatment. The contents of glycyrrhizic acid and liquiritin varied significantly among different water and fertilizer treatments, with the highest levels observed under the W2F2 treatment. Excessive irrigation (W4) and over-fertilization (F4) led to a decrease in these bioactive components. A comprehensive evaluation of the growth characteristics and bioactive components revealed that the ideal irrigation and fertilization parameters were 4000 m³/ha and 610 kg/ha, respectively. These parameters optimized plant development and bioactive component accumulation while ensuring efficient resource use. This study provides scientific evidence for optimizing irrigation and fertilization strategies to enhance licorice yield in arid regions, thereby supporting sustainable agricultural practices and improving economic benefits.

Meta Learning to Rank for Sparsely Supervised Queries

Supervisory signals are a critical resource for training learning to rank models. In many real-world search and retrieval scenarios, these signals may not be readily available or could be costly to obtain for some queries. The examples include domains where labeling requires professional expertise, applications with strong privacy constraints, and user engagement information that are too scarce. We refer to these scenarios as sparsely supervised queries which pose significant challenges to traditional learning to rank models. In this work, we address sparsely supervised queries by proposing a novel meta learning to rank framework which leverages fast learning and adaption capability of meta-learning. The proposed approach accounts for the fact that different queries have different optimal parameters for their rankers, in contrast to traditional learning to rank models which only learn a global ranking model applied to all the queries. In consequence, the proposed method would yield significant advantages especially when new queries are of different characteristics with the training queries. Moreover, the proposed meta learning to rank framework is generic and flexible. We conduct a set of comprehensive experiments on both public datasets and a real-world e-commerce dataset. The results demonstrate that the proposed meta-learning approach can significantly enhance the performance of learning to rank models with sparsely labeled queries.

Enhancing Graph Neural Networks via Memorized Global Information

Graph neural networks (GNNs) have gained significant attention for their impressive results on different graph-based tasks. The essential mechanism of GNNs is the message-passing framework, whereby node representations are aggregated from local neighborhoods. Recently, Transformer-based GNNs have been introduced to learn the long-range dependencies, enhancing performance. However, their quadratic computational complexity, due to the attention computation, has constrained their applicability on large-scale graphs. To address this issue, we propose MGIGNN ( M emorized G lobal I nformation G raph N eural N etwork), an innovative approach that leverages memorized global information to enhance existing GNNs in both transductive and inductive scenarios. Specifically, MGIGNN captures long-range dependencies by identifying and incorporating global similar nodes, which are defined as nodes exhibiting similar features, structural patterns and label information within a graph. To alleviate the computational overhead associated with computing embeddings for all nodes, we introduce an external memory module to facilitate the retrieval of embeddings and optimize performance on large graphs. To enhance the memory-efficiency, MGIGNN selectively retrieves global similar nodes from a small set of candidate nodes. These candidate nodes are selected from the training nodes based on a sparse node selection distribution with a Dirichlet prior. This selecting approach not only reduces the memory size required but also ensures efficient utilization of computational resources. Through comprehensive experiments conducted on ten widely-used and real-world datasets, including seven homogeneous datasets and three heterogeneous datasets, we demonstrate that our MGIGNN can generally improve the performance of existing GNNs on node classification tasks under both inductive and transductive settings.

Transfer Learning Enabled Modeling Paradigm for PVT-aware Circuit Performance Estimation

Designing robust performance models for modern complex digital circuits in the face of rapidly accelerating process variations is a critical yet demanding task. This paper introduces an efficient statistical performance modeling approach for VLSI digital circuits that incurs minimal computational expense. The fundamental concept involves capitalizing on knowledge gained from circuit modeling in one technology node to streamline the modeling process in another. This is achieved by merging previously established statistical models of process technology with a limited set of simulation data from a subsequent process technology through transfer learning. Comprehensive experiments conducted across diverse technology nodes demonstrate that the proposed framework is robust, precise, efficient in data usage, and computationally superior to other cutting-edge performance modeling techniques.

Changes in the Phylogenetic Structure of Alpine Grassland Plant Communities on the Qinghai-Tibetan Plateau with Long-Term Nitrogen Deposition.

Nitrogen (N) deposition rates have notably increased around the world, especially in high-altitude regions like the Qinghai-Tibetan Plateau (QTP). We conducted a six-year comprehensive experiment to simulate nitrogen deposition in an alpine grassland area near Qinghai Lake. Four levels of nitrogen depositions, i.e., 0 (CK), 8 kg N ha-1year-1 (N1), 40 kg N ha-1year-1 (N2), and 72 kg N ha-1year-1 (N3), with three replicates for each N treatment, were tested annually in early May and early July, with the meticulous collection of plant and soil samples during the peak growth period from 15 July to 15 August. We used the null model to evaluate the impact of environmental filtration and interspecific competition on the dynamics of the plant community was assessed based on the level of discrete species affinities within the plant community by constructing a phylogenetic tree. The results showed that the environmental filter was the predominant driver for the change of community's genealogical fabric. The N2 and N3 treatments increased the influence of soil factors on the change of plant community structure. Climatic factors played a crucial role on the change of plant community in the CK grassland area, while soil factors were dominant in the N1- and N3-treated grasslands.

DAPLSR: Data Augmentation Partial Least Squares Regression Model via Manifold Optimization

Traditional Partial Least Squares Regression (PLSR) models frequently underperform when handling data characterized by uneven categories. To address the issue, this paper proposes a Data Augmentation Partial Least Squares Regression (DAPLSR) model via manifold optimization. The DAPLSR model introduces the Synthetic Minority Over-sampling Technique (SMOTE) to increase the number of samples and utilizes the Value Difference Metric (VDM) to select the nearest neighbor samples that closely resemble the original samples for generating synthetic samples. In solving the model, in order to obtain a more accurate numerical solution for PLSR, this paper proposes a manifold optimization method that uses the geometric properties of the constraint space to improve model degradation and optimization. Comprehensive experiments show that the proposed DAPLSR model achieves superior classification performance and outstanding evaluation metrics on various datasets, significantly outperforming existing methods.

Multi‐objective based container placement strategy in CaaS

AbstractIn contrast to a conventional virtual machine (VM), a container is a lightweight virtualization technology. Containers are becoming a prominent technology for cloud services because of their portable, scalable, and flexible deployments, especially in the Internet of Things (IoT), smart devices, and fog and edge computing. It is a type of operating system‐level virtualization in which the kernel allows multiple isolated containers to run independently. Container placement (CP) is a nontrivial problem in Container‐as‐a‐Service (CaaS). CP is mapping to a container over virtual machines (VMs) to execute an application. Designing an efficient CP strategy is complex due to several intertwined challenges. These challenges arise from a diverse spectrum of computing resources, like on‐demand and unpredictable fluctuations of IT resources by multiple tenants. In this article, we propose a modified sum‐based container placement algorithm called a multi‐objective optimization‐based container placement algorithm (MSBCPA). In the proposed algorithm, we have considered two metrics: makespan and monetary costs for optimizing available IT resources. We have conducted comprehensive simulation experiments to validate the effectiveness of the proposed algorithm over the CloudSim 4.0 simulator. The proposed optimization algorithm (MSBCPA) aims to minimize the makespan and the execution monetary costs simultaneously. In the simulation, we found that the execution cost and energy consumption cost reduce by 20% to 30% and achieve the best possible cost‐makespan trade‐offs compared to competing algorithms.

Enhanced electrical and mechanical properties of additively manufactured pure copper with green laser

The effort to improve the additive manufacturing (AM) of copper, which is essential in various industrial sectors, has led to the exploration of green laser powder bed fusion (GL-PBF). By using a high-energy green laser, we overcome the challenges posed by the high reflectivity of copper, which has previously hindered achieving the desired component densities and functionalities through AM. This study uncovers the key role of GL-PBF process parameters on the densification, microstructure, surface roughness, mechanical properties, and electrical conductivity of pure copper parts. Our findings demonstrate that optimizing GL-PBF parameters can achieve copper components with over 99.9 % relative density and 98 % international annealed copper standard (IACS) electrical conductivity. The combination of comprehensive experiments and finite element modeling also reveals how the critical role of defect morphology in affecting electrical conductivity. This work contributes to the broader application of AM technologies, especially for high-reflectivity metals, and also provides new insights into how these defects affect conductivity and should be controlled during the AM process.

Semi-supervised action recognition with dynamic temporal information fusion

The most advanced semi-supervised models available are based on images for innovation, and the use of semi-supervised learning models augmented with temporal data for video-level action recognition still suffers from severe model mismatches, and the models are not sufficiently capable of capturing both local and global information about the action. Secondly the use of constant-threshold pseudo-labeling leads to low utilization of unlabeled data for difficult actions in the early stages of training, poor pseudo-labeling quality and affects recognition accuracy. To make the semi-supervised framework FixMatch more suitable for action recognition, we propose Time-Mixer and Dynamic Threshold, respectively. Time-Mixer explores complementary information between time sequences through the fusion of two-channel temporal context information. Dynamic Threshold utilizes a new core mapping function (Normal Distribution Function) to enhance pseudo-labeling quality. Extensive experiments were conducted on three action recognition datasets (Kinetics-400, UCF-101, and HMDB-51). Comprehensive experiments show that the performance of the semi-supervised model in action recognition improves considerably after using dynamic thresholding and temporal context information fusion, with a 14.4% improvement over the baseline and a 1.8% improvement over the TG (with a labeling rate of 10%) in UCF101, whereas an overall good performance is obtained for DTIF.

BioSAM: Generating SAM Prompts From Superpixel Graph for Biological Instance Segmentation.

Proposal-free instance segmentation methods have significantly advanced the field of biological image analysis. Recently, the Segment Anything Model (SAM) has shown an extraordinary ability to handle challenging instance boundaries. However, directly applying SAM to biological images that contain instances with complex morphologies and dense distributions fails to yield satisfactory results. In this work, we propose BioSAM, a new biological instance segmentation framework generating SAM prompts from a superpixel graph. Specifically, to avoid over-merging, we first generate sufficient superpixels as graph nodes and construct an initialized graph. We then generate initial prompts from each superpixel and aggregate them through a graph neural network (GNN) by predicting the relationship of superpixels to avoid over-segmentation. We employ the SAM encoder embeddings and the SAM-assisted superpixel similarity as new features for the graph to enhance its discrimination capability. With the graph-based prompt aggregation, we utilize the aggregated prompts in SAM to refine the segmentation and generate more accurate instance boundaries. Comprehensive experiments on four representative biological datasets demonstrate that our proposed method outperforms state-of-the-art methods.

Soft-label recover based label-specific features learning

Presently, multi-label classification algorithms are mainly based on positive and negative logical labels, which have achieved good results. However, logical labeling inevitably leads to the label misclassification problem. In addition, missing labels are common in multi-label datasets. Recovering missing labels and constructing soft labels that reflect the mapping relationship between instances and labels is a difficult task. Most existing algorithms can only solve one of these problems. Based on this, this paper proposes a soft-label recover based label-specific features learning (SLR-LSF) to solve the above problems simultaneously. Firstly, the information entropy is used to calculate the confidence matrix between labels, and the membership degree of soft labels is obtained by combining the label density information. Secondly, the membership degree and confidence matrix are combined to construct soft labels, and this process not only solves the problem of missing labels but also obtains soft labels with richer semantic information. Finally, in the process of learning specific label features for soft labels. The local smoothness of the labels learned through stream regularization is complemented by the global label correlation, thus improving the classification performance of the algorithm. To demonstrate the effectiveness of the proposed algorithm, we conduct comprehensive experiments on several datasets.

Improving Crystal Property Prediction from a Multiplex Graph Perspective.

Graph neural networks (GNNs) have proven to be effective tools for the rapid and accurate prediction of crystal properties. While most existing methods focus on enriching representations of crystal structures, they do not deeply explore the characteristics of crystal graphs and leverage their intrinsic information from a data science perspective. In this work, we propose the potential multiplex crystal graph neural network (PMCGNN) for crystal property prediction. Based on the characteristics of crystal graphs, we reconstruct the crystal graph into a multiplex graph that includes two views: a global crystal graph embodying infinite potentials and a local crystal graph capturing local atomic interactions. We employ graph transformers (GTs) and message passing neural networks (MPNNs) architectures to learn the atomic representations of these two perspectives. Specifically, we augment the GT by incorporating positional encodings and structural encodings from the local crystal graph. This approach promotes interaction between the two perspectives, enabling the model to learn both node positional and graph structural information from different viewpoints through an attention mechanism. As a result, it enhances the model's ability to learn crystal representations. We conduct comprehensive experiments on the JARVIS and the Materials Project data sets for evaluation. Results show that PMCGNN presents superior performance in 9 crystal prediction tasks while maintaining reasonable computational expense.