With the advent of 5G, the internet has entered a new video-centric era. From short-video platforms like TikTok to long-video platforms like Bilibili, online video services are reshaping user consumption habits. Adaptive Bitrate (ABR) control is widely recognized as a critical factor influencing Quality of Experience (QoE). Recent learning-based ABR methods have attracted increasing attention. However, most of them rely on limited network trace sets during training and overlook the wide-distribution characteristics of real-world network conditions, resulting in poor generalization in out-of-distribution (OOD) scenarios. To address this limitation, we propose SABR, a training framework that combines behavior cloning (BC) pretraining with reinforcement learning (RL) fine-tuning. We also introduce benchmarks, ABRBench-3G and ABRBench-4G+, which provide wide-coverage training traces and dedicated OOD test sets for assessing robustness to unseen network conditions. Experimental results demonstrate that SABR achieves the best average rank compared with Pensieve, Comyco, and NetLLM across the proposed benchmarks. These results indicate that SABR enables more stable learning across wide distributions and improves generalization to unseen network conditions.

Computational drug repositioning technology can identify potential uses for existing drugs and reduce the time and cost required in the drug development process. How to find appropriate representations of drugs and diseases to predict the associations between the two is the main objective of such tasks. With the emergence of graph neural networks in recent years, researchers learned drugs and diseases via graphs in a bid to improve the prediction accuracy. However, there are three key problems that have not been adequately studied: 1) They usually place drug-disease association graphs, drug-drug similarity graphs, and disease-disease similarity graphs under the same semantic space for learning, which lose the higher-order features of the different graphs. 2) They assign equal weight to each neighbor node when aggregating based on the drug-disease association graph, but the effect and mechanism of a drug in treating different diseases are not consistent. 3) They adopt residual connections to enhance the role of the root node without considering that this operation amplifies the effect of anomalous features. In view of this, we first propose a graph attention network-based spatial decomposition method for drug repositioning. Specifically, we reduce the dimensions of the feature space by spatial decomposition and initialize the drug and disease embedding in the drug-similarity subspace, disease-similarity subspace, and drug-disease association subspace, respectively. The representations of drugs and diseases are jointly captured in the corresponding subspace based on similarity and therapeutic associations. Moreover, the extent of the associations is measured through the graph attention mechanism to explore higher-order neighborhood relationships between drugs and diseases. Finally, we introduce a targeted residual connection for personalized propagation of node features. Experiments on four benchmark datasets show that our proposed architecture outperforms current state-of-the-art approaches.

Offline reinforcement learning (offline RL) endeavors to learn effective policies from a large batch of pre-collected datasets without any costly or dangerous online exploration. Nevertheless, offline RL always suffers from substantial algorithmic extrapolation errors and may fail when bootstrapping from out-of-distribution (OOD) actions or states. In this work, we introduce a practical and effective Bayesian uncertainty weighted optimization (BUWO) to leverage the Bayesian uncertainty to account for the epistemic uncertainty associated with each training sample and penalize the state-action pairs with high uncertainty. We compare BUWO with other prevailing offline RL algorithms on D4RL benchmarks. The experimental results demonstrate that the algorithm can enhance the average reward score by almost 15% without additional computational costs compared to the current state-of-the-art algorithm.

Recent advancements in Swin Transformer-based image super-resolution (SR) methods have demonstrated remarkable progress through efficient modeling capabilities and feature extraction. However, existing approaches still have limitations: in hierarchical feature representation architectures, inter-layer feature propagation efficiency progressively degrades with increasing network depth. For this, the paper proposes an SR model, namely, SwinDPSR, composed of Dense-Parallel Swin Transformer (DPST) blocks. The proposed method innovatively constructs a densely connected parallel Swin Transformer architecture that enhances information flow through multi-level feature reuse mechanisms. Simultaneously, we introduce the Spatial Frequency Block (SFB) that establishes global contextual correlations in the frequency domain branch to precisely compensate for local detail loss. Experiments demonstrate significant improvements over the SwinIR across five benchmarks (Set5, Set14, BSD100, Urban100, and Manga109), achieving an average PSNR gain of 0.28[Formula: see text]dB in [Formula: see text] SR tasks. Theoretical analysis and experimental validation confirm the effectiveness of the parallel architecture, which improves FPS by 6.8% through enhanced computational parallelism. Ablation studies verify the synergistic optimization effect of dense connections and SFB.

Scene text detection is a challenging topic in computer vision, characterized by complex illumination, irregular shape, and arbitrary size. While recent advancements have been made in scene text detection, it remains difficult to simultaneously distinguish nearby text and accommodate irregularly shaped text. Therefore, this paper introduces HPNet, an enhanced text detector, based on the segmentation method that predicts two-scale results. To improve the shape robustness, the Hybrid Attentional Feature Fusion (HAFF) module is integrated into Feature Pyramid Networks (FPN) to dynamically perform feature fusion. Additionally, to distinguish nearby text, the model predicts the text region covering text instances and the text kernel covering the central region of the text. The improved Pixel Aggregation (PA) algorithm is then utilized to guide the expansion from the text kernel to the text region. Experiments on IC15, Total-Text, and CTW1500 validate the effectiveness of these improvements and the superiority of HPNet. Compared with the previous method PSENet for nearby texts, the proposed HPNet has improved inference speed by 63.6% and F-measure metric by 2.6%, 3.7%, and 2.5% on three datasets, respectively.

Efficient and intelligent exploration remains a major challenge in the field of deep reinforcement learning (DRL). Bayesian inference with a distributional representation is usually an effective way to improve the exploration ability of the RL agent. However, when optimizing Bayesian neural networks (BNNs), most algorithms need to specify an explicit parameter distribution such as a multivariate Gaussian distribution. This may reduce the flexibility of model representation and affect the algorithm performance. Therefore, to improve sample efficiency and exploration based on Bayesian methods, we propose a novel implicit posteriori parameter distribution optimization (IPPDO) algorithm. First, we adopt a distributional perspective on the parameter and model it with an implicit distribution, which is approximated by generative models. Each model corresponds to a learned latent space, providing structured stochasticity for each layer in the network. Next, to make it possible to optimize an implicit posteriori parameter distribution, we build an energy-based model (EBM) with value function to represent the implicit distribution which is not constrained by any analytic density function. Then, we design a training algorithm based on amortized Stein variational gradient descent (SVGD) to improve the model learning efficiency. We compare IPPDO with other prevailing DRL algorithms on the OpenAI Gym, MuJoCo, and Box2D platforms. Experiments on various tasks demonstrate that the proposed algorithm can represent the parameter uncertainty implicitly for a learned policy and can consistently outperform competing approaches.

Nonuniform thermal behavior in lithium-ion battery packs can accelerate aging, leading to inconsistent cell performance. If not adequately monitored and managed, this heating can give rise to unwanted side reactions, fires, and explosions, underscoring the criticality of temperature field reconstruction. In recent years, data-driven methods have gained popularity for addressing the temperature field reconstruction problem. However, many existing data-driven approaches require retraining when system parameters change, such as the initial temperature distribution or working conditions. This article presents a deep transfer operator learning method named physics-informed adversarial networks. The model architecture incorporates transformer blocks to capture comprehensive time and space features. Additionally, to enhance interpretability and generalization, the model introduces two effective mechanisms: 1) the integration of thermal partial differential equations to ensure compliance with physical laws; and 2) the application of domain adversarial mechanism in transfer learning to extract domain-invariant feature representations. These mechanisms enable the model to effectively reconstruct the temperature field, even in unencountered scenarios during training. The proposed method is validated under real-world energy storage working conditions, demonstrating superior performance compared to state-of-the-art deep learning methods. Notably, the approach exhibits excellent performance even when confronted with the limited availability of training data.

Drug-target binding affinity prediction plays an important role in the early stages of drug discovery, which can infer the strength of interactions between new drugs and new targets. However, the performance of previous computational models is limited by the following drawbacks. The learning of drug representation relies only on supervised data without considering the information in the molecular graph itself. Moreover, most previous studies tended to design complicated representation learning modules, while uniformity used to measure representation quality is ignored. In this study, we propose GraphCL-DTA, a graph contrastive learning with molecular semantics for drug-target binding affinity prediction. This graph contrastive learning framework replaces the dropout-based data augmentation strategy by performing data augmentation in the embedding space, thereby better preserving the semantic information of the molecular graph. A more essential and effective drug representation can be learned through this graph contrastive framework without additional supervised data. Next, we design a new loss function that can be directly used to adjust the uniformity of drug and target representations. By directly optimizing the uniformity of representations, the representation quality of drugs and targets can be improved. The effectiveness of the above innovative elements is verified on two real datasets, KIBA and Davis. Compared with the GraphDTA model, the relative improvement of the GraphCL-DTA model on the two datasets is 2.7% and 4.5%. The graph contrastive learning framework and uniformity function in the GraphCL-DTA model can be embedded into other computational models as independent modules to improve their generalization capability.

Computational drug repositioning aims to discover new therapeutic diseases for marketed drugs and has the advantages of low cost, short development cycle, and high controllability compared to traditional drug development. The matrix factorization model has become the cornerstone technique for computational drug repositioning due to its ease of implementation and excellent scalability. However, the matrix factorization model uses the inner product operation to represent the association between drugs and diseases, which is lacking in expressive ability. Moreover, the degree of similarity of drugs or diseases could not be implied on their respective latent factor vectors, which is not satisfy the common sense of drug discovery. Therefore, a neural metric factorization model for computational drug repositioning (NMFDR) is proposed in this work. We novelly consider the latent factor vector of drugs and diseases as a point in the high-dimensional coordinate system and propose a generalized euclidean distance to represent the association between drugs and diseases to compensate for the shortcomings of the inner product operation. Furthermore, by embedding multiple drug (disease) metrics information into the encoding space of the latent factor vector, the information about the similarity between drugs (diseases) can be reflected in the distance between latent factor vectors. Finally, we conduct wide analysis experiments on three real datasets to demonstrate the effectiveness of the above improvement points and the superiority of the NMFDR model.

Computational drug repositioning technology is an effective tool to accelerate drug development. Although this technique has been widely used and successful in recent decades, many existing models still suffer from multiple drawbacks such as the massive number of unvalidated drug-disease associations and the inner product. The limitations of these works are mainly due to the following two reasons: firstly, previous works used negative sampling techniques to treat unvalidated drug-disease associations as negative samples, which is invalid in real-world settings; secondly, the inner product cannot fully take into account the feature information contained in the latent factor of drug and disease. In this paper, we propose a novel PUON framework for addressing the above deficiencies, which models the risk estimator of computational drug repositioning only using validated (Positive) and unvalidated (Unlabelled) drug-disease associations without employing negative sampling techniques. The PUON also proposed an Outer Neighborhood-based classifier for modeling the cross-feature information of the latent facotor. For a comprehensive comparison, we considered 6 popular baselines. Extensive experiments in four real-world datasets showed that PUON model achieved the best performance based on 6 evaluation metrics.

The computational drug repositioning aims to discover new uses for marketed drugs, which can accelerate the drug development process and play an important role in the existing drug discovery system. However, the number of validated drug-disease associations is scarce compared to the number of drugs and diseases in the real world. Too few labeled samples will make the classification model unable to learn effective latent factors of drugs, resulting in poor generalization performance. In this work, we propose a multi-task self-supervised learning framework for computational drug repositioning. The framework tackles label sparsity by learning a better drug representation. Specifically, we take the drug-disease association prediction problem as the main task, and the auxiliary task is to use data augmentation strategies and contrast learning to mine the internal relationships of the original drug features, so as to automatically learn a better drug representation without supervised labels. And through joint training, it is ensured that the auxiliary task can improve the prediction accuracy of the main task. More precisely, the auxiliary task improves drug representation and serving as additional regularization to improve generalization. Furthermore, we design a multi-input decoding network to improve the reconstruction ability of the autoencoder model. We evaluate our model using three real-world datasets. The experimental results demonstrate the effectiveness of the multi-task self-supervised learning framework, and its predictive ability is superior to the state-of-the-art model.

The Markov random field (MRF) for stereo matching can be solved using belief propagation (BP). However, the solution space grows significantly with the introduction of high-resolution stereo images and 3D plane labels, making the traditional BP algorithms impractical in inference time and convergence. We present an accurate and efficient hierarchical BP framework using the representation of the image segmentation pyramid (ISP). The pixel-level MRF can be solved by a top-down inference on the ISP. We design a hierarchy of MRF networks using the graph of superpixels at each ISP level. From the highest/image to the lowest/pixel level, the MRF models can be efficiently inferred with constant global guidance using the optimal labels of the previous level. The large texture-less regions can be handled effectively by the MRF model on a high level. The advanced 3D continuous labels and a novel support-points regularization are integrated into our framework for stereo matching. We provide a data-level parallelism implementation which is orders of magnitude faster than the best graph cuts (GC) algorithm. The proposed framework, HBP-ISP, outperforms the best GC algorithm on the Middlebury stereo matching benchmark.

Security evaluation is an essential activity for understanding the risks of industrial networks. Constant changes in networks and updates of security vulnerabilities lead to increasing costs of security testing. To realize the combined exploitation of multiple vulnerabilities in the industrial control network, Knowledge Graph is proposed to be applied in knowledge-driven attack strategy generation in this paper. A graph paradigm of knowledge about vulnerability exploitation is established by integrating and extracting multi-dimensional domain knowledge. By occupying each device-level node, attack strategies based on domain knowledge improve the performance in comprehensive vulnerability exploitation and flexible response. The feasibility of our method is demonstrated through an industrial network example.

The growth of massive generated data from the industrial field results in performance reduction on network bandwidth, real-time response, and communication latency. For distributed industrial automation systems, edge computing suits better than cloud computing from real-time constraints purpose. Furthermore, the integration of OPC UA in edge computing nodes improves interoperability and compatibility. This paper focuses on computation task reconfiguration issues among physically neighboring nodes based on OPC UA. Considering the high real-time requirement in industrial edge computing, a task reconfiguration mechanism is proposed and a real-time workload reallocation algorithm for optimization without any iteration process. Finally, a case study on a triplex reciprocating pump with a multi-class faults simulation model proves that it can improve task execution efficiency without worsening classification accuracy much.

Internet of Things (IoT) has become a new paradigm of communication to reform traditional industries, in which distributed data automatically collected via IoT in a low cost enables many new IT services that were even impossible decades ago. This research reports on an IoT-enabled production management service for crude-oil industry. In practice, even if an optimal management decision is achieved, disruptions, such as possible oil-device failures, inclement weathers and other disturbances, arise frequently and then weaken efficiency and stability of supplement. With the help of IoT, a near-real-time management service comes into being, although the adoption of IoT brings new challenges to management of disruptions. The contributions of this article are as follows: First, a service framework is proposed for refinery which combines MQTT and Azure cloud, enabling reliable data/command delivery. Second, a smart disruption management service system is developed, which consists of monitor and alarm module, disruption management module, and rescheduling procedure module. The rescheduling procedure module takes into account the network of the refinery operations, and is easy to accommodate changes in the refinery configuration for unforeseen disruptions. The experimental results show that the proposed disruption management method balances efficiency and stability compared to traditional methods.

Automatic code generation is a critical step towards flexible manufacturing processes. In the last decade, Model-Driven Engineering is commonly adopted for code generation while requiring tight coupling between the model and its corresponding code, which increases the difficulty of dynamic reconfiguration in the system. To improve the flexibility and efficiency of industrial software design and development processes, Knowledge Graph is proposed to be applied in knowledge-driven code generation process in this paper. The knowledge-driven query system can conduct parameter searching, variable calculation, ontology reasoning, and code invocation to assist code generation. In addition, a structure of domain-specific knowledge graph combined with SQL database and reasoning rules are proposed to improve performance. The feasibility of our method is demonstrated through the dynamic AGV route planning example.

Human pose estimation for multiple people is more challenging than single-person case because of the unknown number of people in an image and occlusion between people. In this paper, we present a generic bottom-up approach for multi-person pose estimation. We introduce limb detection heatmaps as a representation of body joint pairs association, which are simultaneously learned with joint detections. The bottom-up inference is formulated as a set of bipartite graph matching. Our approach can be easily incorporated with any single-person pose estimator that produces joint detection heatmaps with minimal modification. Moreover, we adopt focal loss to address the data imbalance in human pose estimation. Promising results are achieved on the MPII benchmark.

Industrial Cyber-physical systems (ICPSs) are expected to provide effective solutions for improving the operation of many existing industrial manufacturing systems. Wireless sensor networks in the industrial field is classified as low-power and lossy network due to energy constrained devices, the dynamic environment and a high packet loss rate. Energy efficiency and delivery reliability need to be achieved to cope with limited resources and the dynamic environment. In this paper, we propose a cross-layer design CDER to achieve energy efficiency and ensure the reliability of transmission based on 6LoWPAN, considering the routing stability. For purpose of achieving energy conservation, we propose a packet reassembly algorithm to decrease the number of forward packets. Moreover, for stable route establishment to deal with the dynamic environment, an energy balance algorithm considering routing stability is proposed. Extensive simulations show that our algorithm CDER improves the network performance, while achieving energy efficiency and improving routing stability.

One of the goals in the study of infectious disease is to construct a reliable predictive model on the pathogen-host interactome. Conventional methods on the construction of model consider the problem as a binary classification problem. However, most databases only consist of detected interactions and lack of negative results. Thus, as compare to binary classification, this situation is closer to the collaborative filtering problem in nature. In this paper, a commonly used collaborative filtering technique, matrix factorization is applied on the prediction of pathogen-host interaction. However, in matrix factorization, estimation of latent variables is highly dependent on the completeness of the dataset. If the dataset is incomplete, due to the lack of information, estimation of some latent vectors may be infeasible. To relieve this issue, an extension of probabilistic matrix factorization is proposed in this paper. In the extended model, similarities between objects are taken into account as a basis of estimation. Experiment results have shown that when the sparsity increases, as compare to the conventional matrix factorization model and the probabilistic based matrix factorization model, the similarity based probabilistic matrix factorization model has the best goodness of fit and a high prediction accuracy.

The inverse problem of identifying unknown parameters of known structure dynamical biological systems, which are modelled by ordinary differential equations or delay differential equations, from experimental data is treated in this paper. A two stage approach is adopted: first, combine spline theory and Nonlinear Programming (NLP), the parameter estimation problem is formulated as an optimization problem with only algebraic constraints; then, a new differential evolution (DE) algorithm is proposed to find a feasible solution. The approach is designed to handle problem of realistic size with noisy observation data. Three cases are studied to evaluate the performance of the proposed algorithm: two are based on benchmark models with priori-determined structure and parameters; the other one is a particular biological system with unknown model structure. In the last case, only a set of observation data available and in this case a nominal model is adopted for the identification. All the test systems were successfully identified by using a reasonable amount of experimental data within an acceptable computation time. Experimental evaluation reveals that the proposed method is capable of fast estimation on the unknown parameters with good precision.

Knowledge of interactions between proteins is an importance piece of information on the understanding of cellular mechanism. One of the challenges is to quantify how similar the two protein-protein interaction (PPI) networks. To provide a candidate solution of this problem, a distance measure for PPI networks is proposed. This distance measure involves solving an optimisation problem with which the objective function being a weighted sum of topological and biological measure of an alignment respectively. To solve this problem, the projected subgradient method is employed. To illustrate the performance and usage of this distance measure, it is applied to the PPI networks of herpesvirus family. Five herpesviruses: EpsteinBarr virus (EBV), Herpes simplex virus (HSV), Mouse Cytomegalovirus (mCMV), Kaposi's sarcoma-associated herpesvirus (KSHV) and Varicella zoster virus (VZV) are considered in this paper. PPI network distances of the five her-pesviruses are computed and visualized using multidimensional scaling (MDS). Results show that the distance measure can reflect the dissimilarity among organisms. In addition our algorithm can also separate herpesvirus subfamilies given only topological information of the PPI network.

This paper proposes a combinational optimization algorithm extremal optimization (EO) for protein structure alignment based on the contact map overlap (CMO) model. EO is a meta-heuristic algorithm, as genetic algorithm and simulated annealing, but with a local fitness introduced to guide the improvement of the optimization. By exploiting similarity matrix between two contact maps, the results demonstrate that our algorithm is significantly faster and gets better results for most of the test sets.