Implicit graph neural network for deep graph transformation

The traditional maneuver decision-making approaches are highly dependent on accurate and complete situation information, and their decision-making quality becomes poor when opponent information is occasionally missing in complex electromagnetic environments. In order to solve this problem, an autonomous maneuver decision-making approach is developed based on deep reinforcement learning (DRL) architecture. Meanwhile, a Transformer network is integrated into the actor and critic networks, which can find the potential dependency relationships among the time series trajectory data. By using these relationships, the information loss is partially compensated, which leads to maneuvering decisions being more accurate. The issues of limited experience samples, low sampling efficiency, and poor stability in the agent training state appear when the Transformer network is introduced into DRL. To address these issues, the measures of designing an effective decision-making reward, a prioritized sampling method, and a dynamic learning rate adjustment mechanism are proposed. Numerous simulation results show that the proposed approach outperforms the traditional DRL algorithms, with a higher win rate in the case of opponent information loss.

GACT-PPIS: Prediction of protein-protein interaction sites based on graph structure and transformer network

The foreground segmentation method is a crucial first step for many video analysis methods such as action recognition and object tracking. In the past five years, convolutional neural network based foreground segmentation methods have made a great breakthrough. However, most of them pay more attention to stationary cameras and have constrained performance on the pan–tilt–zoom (PTZ) cameras. In this paper, an end-to-end deep features homography transformation and fusion network based foreground segmentation method (HTFnetSeg) is proposed for surveillance videos recorded by PTZ cameras. In the kernel of HTFnetSeg, there is the combination of an unsupervised semantic attention homography estimation network (SAHnet) for frames alignment and a spatial transformed deep features fusion network (STDFFnet) for segmentation. The semantic attention mask in SAHnet reinforces the network to focus on background alignment by reducing the noise that comes from the foreground. STDFFnet is designed to reuse the deep features extracted during the semantic attention mask generation step by aligning the features rather than only the frames, with a spatial transformation technique in order to reduce the algorithm complexity. Additionally, a conservative strategy is proposed for the motion map based post-processing step to further reduce the false positives that are brought by semantic noise. The experiments on both CDnet2014 and Lasiesta show that our method outperforms many state-of-the-art methods, quantitively and qualitatively.

Non-point source pollution from agricultural activities poses a significant threat to water quality by introducing excess nutrients like nitrogen into aquatic ecosystems, leading to issues such as eutrophication and groundwater contamination. In agricultural watersheds, nitrate transport involves intricate physical, chemical, and biological processes influenced by meteorological conditions, hydrological features, and spatial topologies, making accurate short-term predictions challenging. Traditional data-driven deep learning models often fail to incorporate physical constraints and complex spatiotemporal dynamics, limiting their interpretability and predictive accuracy. Here we show a hierarchical transformer and graph neural network model that accurately predicts watershed nitrate concentrations by integrating multi-source data and simulating pollutant migration. The model captures nonlinear multivariate temporal patterns through hierarchical transformers, fuses global meteorological and local hydrological features via neural networks, and models runoff topologies with physically constrained graph neural networks. For predicting the concentration changes of pollutants discharged from watersheds, it outperforms baselines like multi-layer perceptrons, recurrent neural networks, and long short-term memory networks, with state-of-the-art performance in root mean square error, mean absolute error, and R2. Ablation studies confirm the essential roles of multi-source data integration and watershed topological modeling in enhancing performance. This method of directly modeling physical processes by leveraging the characteristics of different neural network architectures opens up a new path for addressing the interpretability problem in neural earth system modeling, apart from the process-guided deep learning and differentiable modelling methods.

In a preliminary exploratory study, our goal was to train deep neural network models to mimic children's and/or adults' gaze behavior in certain social situations to reach this objective. Additionally, we aim to identify potential differences in gaze behavior between these two age groups based on our participants' gaze data. Furthermore, we aimed to assess the practical effectiveness of our adult and children models by deploying them on a Nao robot in real-life settings. To achieve this, we first created two video clips, one animation and one live-action, to depict some social situations. Using an eye-tracking device, we collected eye-tracking data from 24 participants, including 12 children and 12 adults. Then, we utilized deep neural networks, specifically LSTM and Transformer Networks, to analyze and model the gaze patterns of each group of participants. Our results indicate that when the models attempted to predict people's locations (in the next frame), they had an accuracy in the range of 62%-70% with one attempt, which increased by ~20% when attempted twice (i.e. the two highest-ranked predicted labels as outputs). As expected, the result underscores that gaze behavior is not a wholly unique phenomenon. We obtained feedback from 57 new participants to evaluate the robot's functionality. These participants were asked to watch two videos of the robot's performance in each mode and then complete a comprehensive questionnaire. The questionnaire results indicate that the participants expressed satisfaction with the robot's interaction, including its attention, intelligence, and responsiveness to human actions. However, they did not perceive the robot as a social companion comparable to a human. This exploratory study tries to address/show potentials of the social acceptance of robots based on human nonverbal behavioral cues for future research.

Deep learning basics are essential for the visual question answering task since multimodal information is usually complex and multidimensional. Therefore, in this chapter, we present basic information regarding deep learning, covering the following: (1) neural networks, (2) convolutional neural networks, (3) recurrent neural networks and their variants, (4) encoder/decoder structure, (5) attention mechanism, (6) memory networks, (7) transformer networks and BERT, and (8) graph neural networks.

DeLiVoTr: Deep and light-weight voxel transformer for 3D object detection

Graph Neural Networks (GNNs) marked their presence in track reconstruction a few years ago. Initial studies eventually grew into mature pipelines, performing end-to-end track reconstruction for different detectors. Publications describing those efforts usually present the selected GNN architectures for specific reconstruction steps. Therefore, in this study, we would like to contribute to research on which of the general-purpose GNN architectures are especially promising in link prediction of high-energy physics (HEP) data. For this analysis, we compare three graph neural networks: one leveraging only SAGEConv layers and two additionally using Graph Transformer or PointNet networks. We also present achievable metrics for the simplified edge classification task. In addition, we advocate for the use of the ACTS toolkit as a simulation testbed and present a simple VELO-inspired toy detector.

Accurate prediction of land surface temperature (LST) is critical for remote sensing applications, yet remains hindered by in situ data scarcity, limited input variables, and regional variability. To address these limitations, we introduce a three-stage strategic fine-tuning-based transfer learning (SFTL) framework that integrates a large simulated dataset (430 K samples), in situ measurements from the Heihe and Huailai regions in China, and high-resolution imagery from the GF5-02 Visible and Infrared Multispectral Imager (VIMI). The key novelty of this study is the combination of large-scale simulation, an engineered humidity-sensitive feature, and multiple parameter-efficient tuning strategies—full, head, gradual, adapter, and low-rank adaptation (LoRA)—within a unified transfer-learning framework for cross-site LST estimation. In Stage 1, pre-training with 5-fold cross-validation on the simulated dataset produced strong baseline models, including Random Forest (RF), Light Gradient Boosting Machine (LGBM), Deep Neural Network (DNN), Transformer (TrF), and Convolutional Neural Network (CNN). In Stage 2, strategic fine-tuning was conducted under two cross-regional scenarios—Heihe-to-Huailai and Huailai-to-Heihe—and model transfer for tree-based learners. Fine-tuning achieved competitive in-domain performance while materially improving cross-site transfer. When trained on Huailai and tested on Heihe, DNN-gradual attained RMSE 2.89 K (R2 ≈ 0.96); when trained on Heihe and tested on Huailai, TrF-head achieved RMSE 3.34 K (R2 ≈ 0.94). In Stage 3, sensitivity analyses confirmed stability across IQR multipliers of 1.0–1.5, with <1% RMSE variation across models and sites, indicating robustness against outliers. Additionally, application to real GF5-02 VIMI imagery demonstrated that the best SFTL configurations aligned with spatiotemporal in situ observations at both sites, capturing the expected spatial gradients. Overall, the proposed SFTL framework—anchored in cross-validation, strategic fine-tuning, and large-scale simulation—outperforms the widely used Split-Window (SW) algorithm (Huailai: RMSE = 3.64 K; Heihe: RMSE = 4.22 K) as well as direct-training Machine Learning (ML) models, underscoring their limitations in modeling complex regional variability.

Current vehicle trajectory prediction methods often fail to capture the complex spatial and temporal interactions among surrounding vehicles in dynamic environments, leading to suboptimal prediction accuracy. In response to these challenges, we propose a novel Transformer-based model for vehicle trajectory prediction, denoted as the Dynamic Spatial-Temporal Interaction-aware Transformer Network (DSTIFormer). The three main parts of the DSTIFormer architecture are a spatial-temporal embedding module, a spatial-temporal dynamics interaction-aware module, and a trajectory encoderdecoder module. The spatial-temporal embedding module aims to encode vehicular trajectory interactions into latent representations that simultaneously preserve spatial dependencies and temporal features. The second module utilizes a temporal transformer network (TTN) along with a spatial graph transformer network (STN) to capture the dynamic spatial-temporal interactions between vehicles. To make trajectory predictions more accurate, the third module uses a sequence-to-sequence (Seq2Seq) architecture as the encoder and an attention-based decoder layer to effectively extract the long-range sequential dependencies in the process. The model's prediction accuracy with the root mean squared error (RMSE) metric on two publicly available datasets: NGSIM and HighD, and the results demonstrate that our proposed model outperforms several benchmark methods in long-term prediction.

In today's digital economy, digital marketing has become a crucial means for businesses to drive growth and enhance brand exposure. However, with increasing competition, predicting and optimizing advertising effectiveness has become a pivotal component in formulating digital marketing strategies. In order to better understand ad creatives and deeply explore the information within them, this study focuses on integrating visual transformer (VIT) and graph neural network (GNN) methods. Additionally, the study leverages generative adversarial networks (GAN) to enhance the quality of visual features, aiming to achieve visual analysis, exploration, and prediction of advertising effectiveness in digital marketing. This approach begins by employing VIT, an emerging visual transformer technology, to transform image information into high-dimensional feature representations.

ObjectiveIn recent years, with the development of machine learning and deep learning technologies, an increasing number of research works have begun using these technologies for automatic seizure detection in EEG signals. However, existing automatic seizure detection algorithms primarily focus on the features of individual EEG channels and pay less attention to the inter-channel relationships. This results in insufficient extraction of spatiotemporal information from multi-channel EEG data, affecting the final seizure detection performance.MethodsTherefore, this paper proposes an automatic seizure detection method based on the combination of Graph Attention Networks (GAT) and Transformer networks. Specifically, GAT is used as the front end for extracting spatial features, fully leveraging the topological structure of different EEG channels. Meanwhile, the Transformer network is used as the back end to explore temporal relationships and make final decisions based on the states before and after the current moment.ResultsExperiments were conducted on the CHB-MIT and TUH datasets with ten-fold cross-validation. The final seizure detection accuracies on the two datasets were 98.62 and 98.12%, respectively, with the model’s performance surpassing or being comparable to current state-of-the-art models.ConclusionThe proposed hybrid algorithm combines the advantages of two deep learning models, fully exploring the spatiotemporal correlations between EEG channels. Experiments on public datasets demonstrate the effectiveness of this method, significantly advancing the development of automatic seizure detection.

Fault diagnosis plays an important part role in ensure the reliability of integrated navigation, the paper was proposed especially for the integrated navigation which used on ship. How to keep its safety must confront to us. Now most of fault diagnosis are based on traditional methods, Because wavelet transformation can localize signals both in time and frequency, and neural network has the ability of self-organized, self-learning. To make fault diagnosis more effective and can adapt to the development of technology. The paper proposed a more intelligent method which combined wavelet transformation and neural network to enhance the efficiency, the summary procedures were designed as below: we combined the wavelet transformation and neural network in series. After three-layer wavelet decomposition, the coefficients of the third layers are achieved. Then 8 eigenvectors as fault samples to train three-layer RBF neural network, for RBF network is good at classification, after training the network can detect a fault on-line. At the same time, it can classify faults and alarmed the users. Gyro signals are chosen as the simulation inputs, the results prove the method applicability and effectiveness.

One prominent feature of cognitive radio (CR) involves spectrum sensing (SS), which allows licensed primary users to remain unaffected by secondary users’ ability to discover and exploit unoccupied frequency bands. Spectrum sensing enhances the use of spectrum in CR devices, increasing their adaptability and efficiency in wireless communication systems. The rise of wireless equipment and the advent of IoT technologies compound this need for flexibility. Over time, the fixed allocation of frequencies has led to inefficiencies and underutilization as bandwidth needs increase. Deep learning and artificial intelligence have improved spectrum sensing by increasing detection probability of primary users’ presence under noisy environments, enabling cognitive radio systems to respond intelligently to fluctuations in RF environments. This article is concerned with deep learning techniques for spectrum sensing and modulation categorization with CBRT structure, which combines convolutional neural networks (CNNs), bidirectional recurrent neural networks (BRNNs), and transformer networks (TNs) to improve spectrum sensing. CNNs are responsible for performing spectrum feature extraction; BRNNs are used to capture temporal dependencies; and TNs are good at long range dependencies. Better performance for this model is aimed by integrating the three architectures described. In the proposed work, six digital modulation schemes were considered, for spectrum sensing. The sensing of spectrum in this model is performed using the RadioML2016.10B open-source dataset and performance metrics like the Jaccard Index (JI), Fowlkes’s Mallows Index, and F1 Score. Modulation classification has been performed using MIGOU-MOD open-source dataset. The proposed model exhibits good detection probability and low sensing error, unlike other methods at lower SNR.

Hybrid CNN-transformer network for interactive learning of challenging musculoskeletal images