Network Extraction Research Articles

Time Series Forecasting Model Based on the Adapted Transformer Neural Network and FFT-Based Features Extraction

Time Series Forecasting Model Based on the Adapted Transformer Neural Network and FFT-Based Features Extraction

A spatial-temporal hierarchical modeling framework for multi-step ride-hailing demand forecasting

Forecasting urban ride-hailing demand is the main duty in Intelligent Transportation Systems. Accurate predicting can significantly improve the efficiency of regional capacity allocation, promoting energy conversation and emission reduction. Currently, the mainstream approaches utilize advanced deep learning-based methods to model aspects of both temporal and spatial. However, these previous methods overlook the long-term temporal dependence of selection and the hierarchical nature of urban road networks. To this end, we propose a novel Spatial-Temporal Hierarchical Network (STHNet) for urban ride-hailing demand prediction. Specially, The Depthwise Separable Convolution Nerual Networks (DSCNNs) extract spatial features of the road network through channel-wise and point-wise convolution operations, while the Nested Long Short-Term Memory networks (NLSTMs) are used to capture the hierarchical temporal dependencies in sequential data. DSCNNs and NLSTMs are cascaded to form the basic module of the multi-step prediction framework, called Spatial-Temporal Hierarchical block. The block can be easily extended to other spatial-temporal modeling tasks. At the end of the network, we introduce a 3DCNN to learn Spatial-Temporal heterogeneity and integrate information. Furthermore, Teacher Forcing and secondary information are incorporated into STHNet to enhance training efficiency. Extensive experiment are conducted on a Xiamen transportation network with 64 regions shows that STHNet outperforms multiple State-Of-The-Art baselines. The qualitative results underscoring the practical engineering applicability of the proposed model.

Multi-source Semi-supervised Condition Constrained Domain Adaptation for Process Fault Classification

Abstract To address inconsistent feature distributions caused by multiple working conditions in industrial processes and the lack of fault type labels, this paper proposes a Multi-source semi-supervised conditional constraint domain adaptation (MSSCCDA) fault classification method. Data from multiple working conditions are divided into multiple source domains and a target domain. A Convolutional neural network (CNN) extracts features from the multi-source domain data, and a triplet loss method reduces the feature distance between the same categories in different source domains. A semi-supervised conditional constraint domain adaptation method is proposed, which combines adversarial domain adaptation with a limited amount of labeled target domain data to pre-train the feature extractor. During adversarial training, center loss regularization is introduced as a conditional constraint for class feature alignment. The process of optimizing the feature extractor involves jointly leveraging adversarial loss to reduce inter-domain discrepancies between the source and target domains and minimizing class differences within target domain. Experiments conducted on the Tennessee-Eisman (TE) process and industrial three-phase flow (TFF) show that the average accuracy improves to 93% and 93%, respectively. The effectiveness of the proposed method is further validated through a comparison with two traditional domain adaptation approaches and five multi-source domain adaptation techniques.

Does This Have a Particular Meaning? Interactive Pattern Explanation for Network Visualizations.

This paper presents an interactive technique to explain visual patterns in network visualizations to analysts who do not understand these visualizations and who are learning to read them. Learning a visualization requires mastering its visual grammar and decoding information presented through visual marks, graphical encodings, and spatial configurations. To help people learn network visualization designs and extract meaningful information, we introduce the concept of interactive pattern explanation that allows viewers to select an arbitrary area in a visualization, then automatically mines the underlying data patterns, and explains both visual and data patterns present in the viewer's selection. In a qualitative and a quantitative user study with a total of 32 participants, we compare interactive pattern explanations to textual-only and visual-only (cheatsheets) explanations. Our results show that interactive explanations increase learning of i) unfamiliar visualizations, ii) patterns in network science, and iii) the respective network terminology.

DeepTGIN: a novel hybrid multimodal approach using transformers and graph isomorphism networks for protein-ligand binding affinity prediction

Predicting protein-ligand binding affinity is essential for understanding protein-ligand interactions and advancing drug discovery. Recent research has demonstrated the advantages of sequence-based models and graph-based models. In this study, we present a novel hybrid multimodal approach, DeepTGIN, which integrates transformers and graph isomorphism networks to predict protein-ligand binding affinity. DeepTGIN is designed to learn sequence and graph features efficiently. The DeepTGIN model comprises three modules: the data representation module, the encoder module, and the prediction module. The transformer encoder learns sequential features from proteins and protein pockets separately, while the graph isomorphism network extracts graph features from the ligands. To evaluate the performance of DeepTGIN, we compared it with state-of-the-art models using the PDBbind 2016 core set and PDBbind 2013 core set. DeepTGIN outperforms these models in terms of R, RMSE, MAE, SD, and CI metrics. Ablation studies further demonstrate the effectiveness of the ligand features and the encoder module. The code is available at: https://github.com/zhc-moushang/DeepTGIN.Scientific contributionDeepTGIN is a novel hybrid multimodal deep learning model for predict protein-ligand binding affinity. The model combines the Transformer encoder to extract sequence features from protein and protein pocket, while integrating graph isomorphism networks to capture features from the ligand. This model addresses the limitations of existing methods in exploring protein pocket and ligand features.

Design of Multimodal Obstacle Avoidance Algorithm Based on Deep Reinforcement Learning

The navigation obstacle avoidance method based on deep reinforcement learning has stronger adaptability and better performance compared to traditional algorithms in complex unknown dynamic environments, and has been widely developed and applied. However, when using multimodal information input, deep reinforcement learning strategy networks extract features that differ significantly between simulated and real world environments, resulting in poor algorithm output strategies and difficulty in transferring models obtained from simulation training to actual environments. To address the aforementioned issues, this article utilizes image segmentation to narrow the gap in environmental features, integrates multimodal information, and designs a deep reinforcement learning multimodal local obstacle avoidance algorithm, MMSEG-PPO, based on proximal strategy optimization algorithms. The algorithm is then ported to practical environments for deployment and testing. The experiment shows that the algorithm proposed in this article reduces the gap between the simulation environment and the actual environment, and has better performance and generalization when transplanted to the real world environment.

Semi-Autogenous Mill Power Consumption Prediction Based on CACN-LSTM

The semi-autogenous (SAG) mill is crucial equipment in the beneficiation process, and power consumption is a key indicator of its operational status. Due to the complex and variable operating environment, the power consumption of the SAG mill has the characteristics of strong coupling of multiple factors, nonlinearity and uncertainty. In order to effectively extract the features that affect the mill power consumption prediction performance and dynamically adjust the weights of each feature, we propose a hybrid prediction model based on channel attention convolutional network (CACN) and long short-term memory (LSTM). The CACN-based network extracts high-dimensional features of input parameters and dynamically assigns weights to them to better capture the key features that characterize the power consumption of the SAG mill, and the LSTM captures long-term dependencies to enable accurate prediction of SAG mill power consumption. To validate the superiority of the proposed method, actual hourly power consumption data from a SAG mill in the beneficiation plant in Yunnan Province is utilized, and experiments are conducted comparing it with models such as GRU, ARIMA, SVM, LSTM, TCN, CNN-GRU, and CNN-LSTM. Experimental results confirm that the proposed model has better prediction performance than other models, and indicators such as R2 have increased by at least 5%.

Regional-Scale Equidistance Optimizing Method Considering the Equidistance Patterns of Discrete Global Grid Systems

The Discrete Global Grid System (DGGS) provides a foundational framework for the digital Earth, where uniform intercell distances are essential for accurate numerical simulations. However, due to the spherical topology, achieving strictly equidistant spherical grid cells is impractical. Most existing studies have focused on regional scales, which are constrained by data acquisition limitations and render global equidistant optimization algorithms economically infeasible. The equidistant characteristics of cells are influenced by map projections and often exhibit regional variations. In this paper, we analyze these equidistant characteristics and construct an equidistant pattern for an icosahedral hexagonal DGGS. By integrating this pattern into the icosahedral orientation method, we develop a regional-scale equidistant optimization method for DGGS. Experiments on river network extraction in the Yangtze River Basin demonstrate significant improvements: the equidistance of grid cells covering the region increased by over 34.2%, while the accuracy of river network extraction improved by 51.41%. Moreover, this method is extensible to other grid models, enhancing the broader applicability of DGGS.

Robust text-dependent speaker verification system using gender aware Siamese-Triplet Deep Neural Network

ABSTRACT Speaker verification in text-dependent scenarios is critical for high-security applications but faces challenges such as voice quality variations, linguistic diversity, and gender-related pitch differences, which affect authentication accuracy. This paper introduces a Gender-Aware Siamese-Triplet Network-Deep Neural Network (ST-DNN) architecture to address these challenges. The Gender-Aware Network utilizes Convolutional 2D layers with ReLU activation for initial feature extraction, followed by multi-fusion dense skip connections and batch normalization to integrate features across different depths, enhancing discrimination between male and female speakers. A bottleneck layer compresses feature maps to capture gender-related characteristics effectively. For enhanced speaker verification, separate male and female ST-DNN models are used, each incorporating Individual, Siamese, and Triplet Networks. The Individual Network extracts unique utterance characteristics, the Siamese Network compares speech sample pairs for speaker identity, and the Triplet Network ensures closely grouped embeddings of samples from the same speaker, facilitating precise verification. Experimental results on RSR2015 and RedDots Challenge 2016 datasets demonstrate significant improvements, with reductions in Equal Error Rate (EER) ranging from 32.31% to 54.55% for males and 33.73% to 38.98% for females, and reductions in MinDCF from 53.47% to 86.36% and 39.46% to 71.19%, respectively, validating the efficacy of the ST-DNN in real-world applications.

Enhanced PM2.5 prediction in Delhi using a novel optimized STL-CNN-BILSTM-AM hybrid model

Accurate air pollution predictions in urban areas facilitate the implementation of efficient actions to control air pollution and the formulation of strategies to mitigate contamination. This includes establishing an early warning system to notify the public. Creating precise estimates for PM2.5 air pollutants in large cities is a challenging task because of the numerous relevant factors and quick fluctuations. This study introduces a novel hybrid model named STL-CNN-BILSTM-AM. It combines the seasonal-trend decomposition method with LOESS (STL) to simplify learning tasks and increase prediction accuracy for complex, nonlinear time-series data. Convolutional neural networks (CNNs) extract features from decomposed components of PM2.5 and other feature variables, such as pollutants and meteorological variables. Bidirectional long-short-term memory (BILSTM) uses these features to extract temporal relationships, enabling the forecasting of daily PM2.5 levels at four locations in Delhi. This hybrid model uses attention mechanisms to extract the most significant information, as well as Bayesian optimization to tune the hyperparameters. The suggested model greatly improved performance in all four regions used in this study, as evidenced by the findings. We compared it with the CNN-BILSTM, BILSTM, LSTM, and CNN models, and the suggested model outperformed the state-of-the-art models by utilizing STL decomposition components and other features. The overall results show that the STL-CNN-BILSTM-AM is better at predicting air quality, especially the concentration of PM2.5 in cities when the data has a high seasonal trend and is complex.Graphical

Interface-aware molecular generative framework for protein–protein interaction modulators

Protein–protein interactions (PPIs) play a crucial role in numerous biochemical and biological processes. Although several structure-based molecular generative models have been developed, PPI interfaces and compounds targeting PPIs exhibit distinct physicochemical properties compared to traditional binding pockets and small-molecule drugs. As a result, generating compounds that effectively target PPIs, particularly by considering PPI complexes or interface hotspot residues, remains a significant challenge. In this work, we constructed a comprehensive dataset of PPI interfaces with active and inactive compound pairs. Based on this, we propose a novel molecular generative framework tailored to PPI interfaces, named GENiPPI. Our evaluation demonstrates that GENiPPI captures the implicit relationships between the PPI interfaces and the active molecules, and can generate novel compounds that target these interfaces. Moreover, GENiPPI can generate structurally diverse novel compounds with limited PPI interface modulators. To the best of our knowledge, this is the first exploration of a structure-based molecular generative model focused on PPI interfaces, which could facilitate the design of PPI modulators. The PPI interface-based molecular generative model enriches the existing landscape of structure-based (pocket/interface) molecular generative model.Scientific contributionThis study introduces GENiPPI, a protein-protein interaction (PPI) interface-aware molecular generative framework. The framework first employs Graph Attention Networks to capture atomic-level interaction features at the protein complex interface. Subsequently, Convolutional Neural Networks extract compound representations in voxel and electron density spaces. These features are integrated into a Conditional Wasserstein Generative AdversarialNetwork, which trains the model to generate compound representations targeting PPI interfaces. GENiPPI effectively captures the relationship between PPI interfaces and active/inactive compounds. Furthermore, in fewshot molecular generation, GENiPPI successfully generates compounds comparable to known disruptors. GENiPPI provides an efficient tool for structure-based design of PPI modulators.

Accurate detection of arbitrary ship directions using SAR based on RTMDet

ABSTRACT Ship detection, an important branch of Synthetic Aperture Radar (SAR) images interpretation, is crucial for various maritime reconnaissance and surveillance but remains challenging due to background interference, dense ship alignments and high aspect ratios. To address these challenges, this letter proposes an arbitrary direction detection method based on Real-Time Models for object Detection (RTMDet), which is one of the fastest and most accurate single-stage detection detectors. Firstly, an attention selection module is designed to replace traditional full convolution method to achieve more parameter reduction; Secondly, in order to extract more key ship feature information and accelerate the network extraction, we construct a lightweight multi-scale feature pyramid network; Thirdly, a novel loss function is introduced to enhance the detection of rotated bounding box by addressing the problem of bounding box discontinuities and lack of scale invariance. The proposed method achieves state-of-the-art detection accuracy on two the high-resolution SAR image datasets for arbitrary ship detection including HRSID (85.6% AP50) and RSDD (90.8% AP50), while demonstrating significant accuracy improvements in the complex inshore scenes.

Spectral Data-Driven Prediction of Soil Properties Using LSTM-CNN-Attention Model

Accurate prediction of soil properties is essential for sustainable land management and precision agriculture. This study presents an LSTM-CNN-Attention model that integrates temporal and spatial feature extraction with attention mechanisms to improve predictive accuracy. Utilizing the LUCAS soil dataset, the model analyzes spectral data to estimate key soil properties, including organic carbon (OC), nitrogen (N), calcium carbonate (CaCO3), and pH (in H2O). The Long Short-Term Memory (LSTM) component captures temporal dependencies, the Convolutional Neural Network (CNN) extracts spatial features, and the attention mechanism highlights critical information within the data. Experimental results show that the proposed model achieves excellent prediction performance, with coefficient of determination (R2) values of 0.949 (OC), 0.916 (N), 0.943 (CaCO3), and 0.926 (pH), along with corresponding ratio of percent deviation (RPD) values of 3.940, 3.737, 5.377, and 3.352. Both R2 and RPD values exceed those of traditional machine learning models, such as partial least squares regression (PLSR), support vector machine regression (SVR), and random forest (RF), as well as deep learning models like CNN-LSTM and Gated Recurrent Unit (GRU). Additionally, the proposed model outperforms S-AlexNet in effectively capturing temporal and spatial patterns. These findings emphasize the potential of the proposed model to significantly enhance the accuracy and reliability of soil property predictions by capturing both temporal and spatial patterns effectively.

MFF-Net: A Lightweight Multi-Frequency Network for Measuring Heart Rhythm from Facial Videos.

Remote photo-plethysmography (rPPG) is a useful camera-based health motioning method that can measure the heart rhythm from facial videos. Many well-established deep learning models can provide highly accurate and robust results in measuring heart rate (HR) and heart rate variability (HRV). However, these methods are unable to effectively eliminate illumination variation and motion artifact disturbances, and their substantial computational resource requirements significantly limit their applicability in real-world scenarios. Hence, we propose a lightweight multi-frequency network named MFF-Net to measure heart rhythm via facial videos in a short time. Firstly, we propose a multi-frequency mode signal fusion (MFF) mechanism, which can separate the characteristics of different modes of the original rPPG signals and send them to a processor with independent parameters, helping the network recover blood volume pulse (BVP) signals accurately under a complex noise environment. In addition, in order to help the network extract the characteristics of different modal signals effectively, we designed a temporal multiscale convolution module (TMSC-module) and spectrum self-attention module (SSA-module). The TMSC-module can expand the receptive field of the signal-refining network, obtain more abundant multiscale information, and transmit it to the signal reconstruction network. The SSA-module can help a signal reconstruction network locate the obvious inferior parts in the reconstruction process so as to make better decisions when merging multi-dimensional signals. Finally, in order to solve the over-fitting phenomenon that easily occurs in the network, we propose an over-fitting sampling training scheme to further improve the fitting ability of the network. Comprehensive experiments were conducted on three benchmark datasets, and we estimated HR and HRV based on the BVP signals derived by MFF-Net. Compared with state-of-the-art methods, our approach achieves better performance both on HR and HRV estimation with lower computational burden. We can conclude that the proposed MFF-Net has the opportunity to be applied in many real-world scenarios.

Budget-Constrained Ego Network Extraction With Maximized Willingness

Budget-Constrained Ego Network Extraction With Maximized Willingness

Born Connected: Do Infants Already Have Adult-Like Multi-Scale Connectivity Networks?

The human brain undergoes remarkable development with the first six postnatal months witnessing the most dramatic structural and functional changes, making this period critical for in-depth research. rsfMRI studies have identified intrinsic connectivity networks (ICNs), including the default mode network, in infants. Although early formation of these networks has been suggested, the specific identification and number of ICNs reported in infants vary widely, leading to inconclusive findings. In adults, ICNs have provided valuable insights into brain function, spanning various mental states and disorders. A recent study analyzed data from over 100,000 subjects and generated a template of 105 multi-scale ICNs enhancing replicability and generalizability across studies. Yet, the presence of these ICNs in infants has not been investigated. This study addresses this significant gap by evaluating the presence of these multi-scale ICNs in infants, offering critical insight into the early stages of brain development and establishing a baseline for longitudinal studies. To accomplish this goal, we employ two sets of analyses. First, we employ a fully data-driven approach to investigate the presence of these ICNs from infant data itself. Towards this aim, we also introduce burst independent component analysis (bICA), which provides reliable and unbiased network identification. The results reveal the presence of these multi-scale ICNs in infants, showing a high correlation with the template (rho > 0.5), highlighting the potential for longitudinal continuity in such studies. We next demonstrate that reference-informed ICA-based techniques can reliably estimate these ICNs in infants, highlighting the feasibility of leveraging the NeuroMark framework for robust brain network extraction. This approach not only enhances cross-study comparisons across lifespans but also facilitates the study of brain changes across different age ranges. In summary, our study highlights the novel discovery that the infant brain already possesses ICNs that are widely observed in older cohorts.

A novel model for mapping soil organic matter: Integrating temporal and spatial characteristics

Mapping the spatial distribution of soil organic matter (SOM) content is crucial for land management decisions, yet its accurate mapping faces challenges due to complex soil-environment relationships and temporal feature capture limitations in machine learning models. This study focuses on the typical black soil region in Northeast China, specifically using Youyi Farm as the main research area and Heshan Farm as the transfer research area. A novel approach is proposed that combines the CNN-LSTM model with a Cosine Annealing Warm Restarts learning rate (CNN-LSTM-CAWR) to enhance the accuracy of SOM mapping. In this model, the Convolutional Neural Network (CNN) extracts spatial context features from static variables (e.g., climate and terrain variables), while the Long Short-Term Memory (LSTM) network captures temporal features from dynamic variables (e.g., Sentinel-2 time series from April to October). The incorporation of the CAWR learning rate helps alleviate overfitting issues. Comparing the CNN-LSTM model, CNN model, and traditional RF model, the results show that the CNN-LSTM-CAWR model achieves the highest accuracy within research Area 1 (R2 = 0.64, RMSE = 0.54 %) and maintains strong performance in the transfer research area (R2 = 0.60, RMSE = 0.57 %). CNN-LSTM-CAWR demonstrates faster convergence, thereby improving mapping precision and effectively utilizing temporal information from features to enhance overall model performance. This study underscores the significant potential of the hybrid CNN-LSTM with CAWR model, highlighting the valuable information for SOM mapping contained within Sentinel-2 time series data.

TF-BAPred: A Universal Bioactive Peptide Predictor Integrating Multiple Feature Representations

Bioactive peptides play essential roles in various biological processes and hold significant therapeutic potential. However, predicting the functions of these peptides is challenging due to their diversity and complexity. Here, we develop TF-BAPred, a framework for universal peptide prediction incorporating multiple feature representations. TF-BAPred feeds original peptide sequences into three parallel modules: a novel feature proposed in this study called FVG extracts the global features of each peptide sequence; an automatic feature recognition module based on a temporal convolutional network extracts the temporal features; and a module integrates multiple widely used features such as AAC, DPC, BPF, RSM, and CKSAAGP. In particular, FVG constructs a fixed-size vector graph to represent the global pattern by capturing the topological structure between amino acids. We evaluated the performance of TF-BAPred and other peptide predictors on different types of peptides, including anticancer peptides, antimicrobial peptides, and cell-penetrating peptides. The benchmarking tests demonstrate that TF-BAPred displays strong generalization and robustness in predicting various types of peptide sequences, highlighting its potential for applications in biomedical engineering.

RNDiff: Rainfall nowcasting with Condition Diffusion Model

The Diffusion Models are widely used in image generation because they can generate high-quality and realistic samples. In contrast, generative adversarial networks (GANs) and variational autoencoders (VAEs) have some limitations in terms of image quality. We introduce a diffusion model to the precipitation forecasting task and propose a short-term precipitation nowcasting with condition diffusion model based on historical observational data, which is referred to as Rainfall nowcasting with Condition Diffusion Model(RNDiff). By incorporating an additional conditional decoder module in the denoising process, RNDiff achieves end-to-end conditional rainfall prediction. RNDiff is composed of two networks: a denoising network and a conditional encoder network. The conditional network is composed of multiple independent UNet networks. These networks extract conditional feature maps at different resolutions, providing accurate conditional information that guides the diffusion model for conditional generation. RNDiff surpasses GANs in terms of prediction accuracy, although it requires more computational resources. The RNDiff model exhibits higher stability and efficiency during training than GANs-based approaches, and generates high-quality precipitation distribution samples that better reflect future actual precipitation conditions. Compared to the current state-of-the-art GAN-based methods, our proposed approach achieves significant improvements on key evaluation metrics. Specifically, our method leads to improvements in the CSI, HSS, and FSS, which are increased by around 8%, 5%, and 6%, respectively. The experiment fully verified the advantages and potential of RNdiff in precipitation forecasting and provided new insights for improving rainfall forecasting. Our project is open source and available on GitHub at: https://github.com/ybu-lxd/RNDiff.

Exact Time-Dependent Thermodynamic Relationships for a Brownian Particle Navigating Complex Networks

The thermodynamic characteristics of systems driven out of equilibrium are examined for M Brownian ratchets organized in a complex network. The precise time-dependent solution reveals that the entropy S, entropy production ep(t), and entropy extraction hd(t) of the system in complex networks increase with system size, which is plausible as these thermodynamic quantities display extensive properties. In other words, as the number of lattice sites increases, the entropy S, entropy production ep(t), and entropy extraction hd(t) increase, demonstrating that these complex networks cannot be reduced to the corresponding one-dimensional lattice. Conversely, the rates for thermodynamic quantities such as velocity V, entropy production rate e˙p(t), and entropy extraction rate h˙ d(t) become independent of the network size in the long-term limit. The exact analytic results also indicate that the free energy decreases with system size. The model system is further analyzed by incorporating heat transfer via kinetic energy. Since heat exchange via kinetic energy does affect the energy extraction rate, the heat dumped to the cold reservoirs also contributes to internal entropy production. Consequently, such systems exhibit a higher degree of irreversibility. The thermodynamic features of a system operating between hot and cold baths are also compared and contrasted with a system functioning in a heat bath where temperature varies linearly along the reaction coordinate. Regardless of the network arrangements, the entropy, entropy production, and extraction rates are significantly larger for the linearly varying temperature case than for a system operating between hot and cold baths.