Latest Issue
- Abstract：Reinforcement learning (RL) has become a dominant decision-making paradigm and has achieved notable success in many real-world applications. Notably, deep neural networks play a crucial role in unlocking RL’s potential in large-scale decision-making tasks. Inspired by current major successes of Transformer in natural language processing and computer vision, numerous bottlenecks have been overcome by combining Transformer with RL for decision-making. This paper presents a multiangle systematic survey of various Transformer-based RL (TransRL) models applied in decision-making tasks, including basic models, advanced algorithms, representative implementation instances, applications, and known challenges. Our work aims to provide insights into problems that inherently arise with the current RL approaches, and examines how we can address them with better TransRL models. To our knowledge, we are the first to present a comprehensive review of the recent Transformers research developments in RL for decision-making. We hope this survey provides a comprehensive review of TransRL models and also inspires the RL community in its pursuit of future directions. Finally, to keep track of the rapid TransRL developments in the decision-making domains, we summarize the latest relevant papers and their open-source implementations at https://github.com/williamyuanv0/Transformer-in-Reinforcement-Learning-for-Decision-Making-A-Survey.Keywords：Transformer;Reinforcement learning;Decision-making;Deep neural network;Multi-Agent Reinforcement Learning;Meta reinforcement learning
- 10
- |
- 9
- |
- 0
Published：2023-12-21 - Abstract：Large-scale deep learning (DL) models are trained distributedly due to memory and computing resource limitations. Few existing strategy generation approaches take optimal memory minimization as the objective. To fill this gap, we propose a novel algorithm that generates optimal parallelism strategies with the constraint of minimal memory redundancy. We propose a novel Redundant Memory Cost Model (RMCM) to calculate the memory overhead of each operator in a given parallel strategy. To generate the optimal parallelism strategy, we formulate the parallelism strategy searching problem into an integer linear programming problem and use an efficient solver to find minimal-memory intra-operator parallelism strategies. Furthermore, the proposed algorithm has been extended and implemented in a multi-dimensional parallel training framework and is characterized by the ability of high throughput and minimal memory redundancy. Experimental results demonstrate that our approach achieves significant memory savings of up to 67% compared to the latest Megatron-LM strategies, and has a similar throughput. The principal contribution of the present research lies in its provision of a novel algorithm that optimizes parallelism strategies, reducing memory redundancy in large-scale DL models. In conclusion, our paper introduces a memory-efficient algorithm for generating parallelism strategies, surpassing existing strategies in reducing memory requirements.Keywords：Deep learning;Automatic parallelism;Minimal memory redundancy
- 3
- |
- 7
- |
- 0
Published：2023-12-21 - Abstract：In this paper, we investigate the impact of network topology characteristics on flocking fragmentation for the multi-robot system under a multi-hop and lossy ad hoc network, including the network’s hop count features and information’s successful transmission probability (STP). More specifically, we first propose a distributed communication-calculation-execution protocol to describe the practical interaction and control process in the ad hoc network based multi-robot system, where the flocking control is realized by a discrete-time Olfati-Saber model incorporating STP-related variables. Then, we develop a fragmentation prediction model (FPM) to formulate the impact of hop count features on fragmentation for specific flocking scenarios. This model identifies the critical system and network features that are associated with fragmentation. Further considering general flocking scenarios affected by both hop count features and STP, we formulate the flocking fragmentation probability (FFP) by a data fitting model based on the back propagation neural network, whose input is extracted from the FPM. The FFP formulation quantifies the impact of key network topology characteristics on fragmentation phenomena. Simulation results verify the effectiveness and accuracy of the proposed prediction model and FFP formulation, and several guidelines for constructing the multi-robot ad hoc network are also concluded.Keywords：Multi-robot flocking;Flocking fragmentation probability;Fragmentation prediction;Multi-robot communication networks
- 21
- |
- 10
- |
- 0
Published：2023-12-21 - Abstract：Phone number recycling (PNR) refers to the event wherein a mobile operator collects a disconnected number and reassigns it to a new owner. It has posed a threat to the reliability of the existing authentication solution for e-commerce platforms. Specifically, a new owner of a reassigned number can access the application account that the number is associated with, and may perform fraudulent activities. Existing solutions that employ a reassigned number database from mobile operators are costly for e-commerce platforms with large-scale users. Thus, alternative solutions that depend on only the information of the applications are imperative. In this work, we study the problem of detecting accounts that have been compromised owing to the reassignment of phone numbers. Our analysis on Meituan’s real-world dataset shows that compromised accounts have unique statistical features and temporal patterns. Based on the observations, we propose a novel model called Temporal pattern and Statistical feature Fusion model (TSF) to tackle the problem, which integrates a Temporal Pattern Encoder and a Statistical Feature Encoder to capture behavioral evolutionary interaction and significant operation features. Extensive experiments on the Meituan and IEEE-CIS datasets show that TSF significantly outperforms the compared baselines, which demonstrates its effectiveness in detecting compromised accounts due to reassigned numbers.Keywords：Phone number recycling;Neural networks;E-commerce;Fraud detection
- 14
- |
- 26
- |
- 0
Published：2023-12-21 - Abstract：This paper uses the semi-tensor product (STP) of matrices and adopts algebraic methods to study the controllability, reachability, and stabilizability of extended finite state machines (EFSMs). Firstly, we construct the bilinear dynamic system model of the EFSM, laying the foundation for further research. Secondly, combined with this bilinear dynamic system model, we propose theorems for the controllability, reachability, and stabilizability of the bilinear dynamic system model of the EFSM. Finally, we design an algorithm to determine the controllability and stabilizability of the EFSM. The correctness of the main results was verified through examples.Keywords：Semi-tensor product;STP approach;STP method;Matrix approach;Algebraic method;Finite-valued systems
- 8
- |
- 29
- |
- 0
Published：2023-12-01 - Abstract：IoT devices are becoming increasingly ubiquitous, and their adoption is growing at an exponential rate. However, they are vulnerable to security breaches, and traditional security mechanisms are not enough to protect them. The massive amounts of data generated by IoT devices can be easily manipulated or stolen, posing significant privacy concerns. This paper is to provide a comprehensive overview of the integration of blockchain and IoT technologies and their potential to enhance the security and privacy of IoT systems. The paper examines various security issues and vulnerabilities in IoT and explores how blockchain-based solutions can be used to address them. It provides insights into the various security issues and vulnerabilities in IoT and explores how blockchain can be used to enhance security and privacy. The paper also discusses the potential applications of blockchain-based IoT systems in various sectors, such as healthcare, transportation, and supply chain management. The paper reveals that the integration of blockchain and IoT has the potential to enhance the security, privacy, and trustworthiness of IoT systems. The multi-layered architecture of B-IoT, consisting of perception, network, data processing, and application layers, provides a comprehensive framework for the integration of these technologies. The study identified various security solutions for B-IoT, including smart contracts, decentralized control, immutable data storage, identity, and access management, and consensus mechanisms. The study also discussed the challenges and future research directions in the field of B-IoT.Keywords：Blockchain;Internet of Things;B-IoT;Security;Scalability;Privacy
- 16
- |
- 13
- |
- 0
Published：2023-12-01 - Abstract：A novel separation identification strategy for the neural fuzzy Wiener-Hammerstein system using hybrid signals was developed in this study. The Wiener-Hammerstein system is described by a model consisting of two linear dynamic elements with a static nonlinear element in between. The static nonlinear element is modeled by a neural fuzzy network (NFN) and the two linear dynamic elements are modeled by an autoregressive exogenous (ARX) model and an autoregressive model (AR), respectively. When the system input is Gaussian signals, the correlation technique is used to decouple the identification of the two linear dynamic elements from the nonlinear element. First, based on the input-output of Gaussian signals, the correlation analysis technique is used to identify the input linear element and output linear element, which addresses the problem that the intermediate variable information cannot be measured in the identified Wiener-Hammerstein system. Then, zero-pole match method is adopted to separate the parameters of the two linear elements. Furthermore, the recursive least squares technique is used for identifying the nonlinear element based on the input-output of random signals, which avoids the impact of output noise. The feasibility of the presented identification technique is demonstrated by an illustrative simulation example and a practical nonlinear process. The simulation results show that the proposed strategy can obtain higher identification precision than existing identification algorithms.Keywords：Wiener-Hammerstein system;Neural fuzzy network;Correlation technique;Hybrid signals;Identification strategy
- 2
- |
- 12
- |
- 0
Published：2023-12-01 - Abstract：Temporal action localization (TAL) is a task of detecting the start and end times of action instances and classifying them in an untrimmed video. As the number of action categories per video increases, existing weakly supervised temporal action localization (W-TAL) methods with only video-level labels cannot provide sufficient supervision. Single-frame supervision has attracted the interest of researchers. Existing paradigms model single-frame annotations from the perspective of video snippet sequences, neglect action discrimination of annotated frames, and do not pay sufficient attention to their correlations in the same category. Considering a category, the annotated frames exhibit distinctive appearance characteristics or clear action patterns. Thus, a novel method to enhance action discrimination via category-specific frame clustering for W-TAL is proposed. Specifically, the K-means clustering algorithm is employed to aggregate the annotated discriminative frames of the same category, which are regarded as exemplars to exhibit the characteristics of the action category. Then, the class activation scores are obtained by calculating the similarities between a frame and exemplars of various categories. Category-specific representation modelling can provide complimentary guidance to snippet sequence modelling in the mainline. As a result, a convex combination fusion mechanism is presented for annotated frames and snippet sequences to enhance the consistency properties of action discrimination, which can generate a robust class activation sequence for precise action classification and localization. Due to the supplementary guidance of action discriminative enhancement for video snippet sequences, our method outperforms existing single-frame annotation-based methods. Experiments conducted on three datasets THUMOS14, GTEA and BEOID show that our method achieves high localization performance compared with state-of-the-art methods.Keywords：Weakly supervised;Temporal action localization;Single-frame annotation;Category specific;Action discrimination Manual;Word template
- 2
- |
- 12
- |
- 0
Published：2023-12-01 - Abstract：In this paper we present the design and realization of a tunable dual band wireless power transfer (TDB-WPT) coupled resonator system. The frequency response of the tunable band can be controlled using a surface mounted (SMD) varactor. The transmitter (Tx) and the receiver (Rx) circuits are symmetric. The top layer contains a feed line with an impedance of 50 Ω. Two identical half rings defected ground structures (HR-DGS) are loaded on the bottom using a varactor diode. We propose a solution for restricted WPT systems working at a single band application according to the operating frequency. The effects of geometry, orientation, relative distance, and misalignments on the coupling coefficients were studied. To validate the simulation results, the proposed TDB-WPT system was fabricated and tested. The system occupies a space of 40 × 40 mm2. It can deliver power to the receiver with an average coupling efficiency of 98% at the tuned band from 817 to 1018 MHz and an efficiency of 95 % at a fixed band of 1.6 GHz at a significant transmission distance of 22 mm. The results of the measurements accord well with those of an equivalent model and the simulation.Keywords：Defected ground structure (DGS);Surface mounted (SMD);Tunable dual band wireless power transfer (TDB-WPT);Varactor
- 3
- |
- 11
- |
- 0
Published：2023-12-01 -
Securing multi-chain consensus against diverse miner behavior attacks in blockchain networks Enhanced Publication
Abstract：As cross-chain technologies enable interactions among different blockchains (hereinafter "chains"), multi-chain consensus is becoming increasingly important in blockchain networks. However, more attention has been paid to single-chain consensus schemes. Multi-chain consensus schemes with trusted miner participation have not been considered, thus offering opportunities for malicious users to launch diverse miner behavior (DMB) attacks on different chains. DMB attackers can be friendly in the consensus process on some chains, called mask chains, to enhance their trust value, while on others, called kill chains, they engage in destructive behavior on the network. In this paper, we propose a multi-chain consensus scheme named Proof-of-DiscTrust (PoDT) to defend against DMB attacks. The distinctive trust idea (DiscTrust) is introduced to evaluate the trust value of each user across different chains. The trustworthiness of a user is split into local and global trust values. A dynamic behavior prediction scheme is designed to enforce DiscTrust to prevent an intensive DMB attacker from maintaining strong trust by alternately creating true or false blocks on the kill chain. Based on this, three trusted miner selection algorithms for multi-chains can be implemented to select network miners, chain miners, and chain miner leaders, respectively. Experimental results show that PoDT is secure against DMB attacks and more effective than traditional consensus schemes in multi-chain environments.Keywords：Blockchain;Cross-chain;Trust mechanism;Multi-chain consensus- 4
- |
- 11
- |
- 0
Published：2023-12-01 - Abstract：With the development of satellite miniaturization and remote sensing, the establishment of microsatellite constellations is an inevitable trend. Due to their limited size, weight, and power (SWaP), spaceborne storage systems with excellent scalability, performance and reliability are still one of the technical bottlenecks of remote sensing microsatellites. Based on the commercial off-the-shelf (COTS) field-programmable gate array (FPGA) and memory devices, a spaceborne advanced storage system (SASS) is proposed in this paper. This work provides a dynamic programming, queue scheduling MIMO cache technique and a high-speed, high-reliability NAND Flash controller for multiple microsatellite payload data. Experimental results show that SASS has outstanding scalability with a maximum write rate of 2429 Mbps and preserves at least 78.53% of the performance when a single NAND Flash fails. The scheduling techniques effectively shorten the data scheduling time, and the data remapping method of the NAND Flash controller reduces the bit error rate (BER) by at least 37.8%.Keywords：Microsatellite;Spaceborne advanced storage system;Scalability;Performance;Reliability
- 4
- |
- 24
- |
- 0
Published：2023-12-01 - Abstract：Low-rank matrix decomposition with first-order total variation (TV) regularization exhibits excellent performance in the exploration of image structure. Taking advantage of its excellent performance in image denoising, we apply it to improve the robustness of deep neural networks. However, although total variation regularization can improve the robustness of the model, it reduces the accuracy of normal samples due to its over-smoothing. In our work, we developed a new low-rank matrix recovery model that incorporates total generalized variation (TGV) regularization into the reweighted low-rank matrix recovery model called LRTGV. In the proposed model, TGV is used to better reconstruct texture information without over-smoothing. The reweighted nuclear norm and L1-norm can enhance the global structure information. Thus, the proposed LRTGV can destroy the structure of adversarial noise while re-enhancing the global structure and local texture of the image. To solve the challenging optimal model issue, we propose an algorithm based on the alternating directions method of multipliers. Experimental results show that the proposed algorithm has a certain defense capability against black-box attacks, and outperforms state-of-the-art low-rank matrix recovery methods in image restoration.Keywords：Total generalized variation;Low-rank matrix;Alternating direction method of the multiplier;Adversarial example
- 2
- |
- 14
- |
- 0
Published：2023-12-01 - Abstract：Electronic healthcare system can offer the convenience but faces the risk of data forgery and information leakage. To solve these issues, we propose an identity-based searchable attribute signcryption in lattice for blockchain-based medical system (BCMSL-IDSASC). The BCMSL-IDSASC achieves the decentralization and anti-quantum security in the blockchain environment, and it provides the fine-grained access control and searchability. Furthermore, the smart contracts are used to replace traditional trusted third parties and interplanetary file system (IPFS) is used for ciphertext storage to alleviate the storage pressure on blockchain. Compared to other schemes, BCMSL-IDSASC offers shorter key size, smaller storage requirement, and lower computation cost. It contributes to secure and efficient management of medical data and can protect the patient privacy and ensure the integrity of electronic healthcare systems.Keywords：Blockchain;Identity-based searchable attribute signcryption;Distributed storage;NTRU lattice
- 2
- |
- 14
- |
- 0
Published：2023-12-01 - Abstract：In this paper, a low-profile dual-broadband dual-circularly-polarized (dual-CP) reflectarray (RA) is proposed and demonstrated, supporting independent beamforming for right-/left-handed CP waves at both K- and Ka-bands. Such functionality is achieved by incorporating multi-layered phase shifting elements individually operating in the K- and Ka-bands, which are then interleaved in a shared aperture, resulting in a cell thickness of only about 0.1λL. By rotating the designed K- and Ka-band elements around their own geometrical centers, the dual-CP waves in each band can be modulated separately. For reducing the overall profile, planar K-/Ka-band dual-CP feeds with a broad bandwidth are designed based on the magnetoelectric dipoles and multi-branch hybrid couplers. The planar feeds achieve bandwidths of about 32% and 26% at K- and Ka-bands with reflection magnitudes below -13 dB, an axial ratio smaller than 2 dB, and a gain variation of less than 1 dB. A proof-of-concept dual-band dual-CP RA integrated with the planar feeds is fabricated and characterized, which is capable of generating asymmetrically distributed dual-band dual-CP beams. The measured peak gain values of the beams are around 24.3 and 27.3 dBic, with joint gain variation < 1 dB and axial ratio < 2 dB bandwidths wider than 20.6% and 14.6% at the lower-band and higher-band, respectively. The demonstrated dual-broadband dual-CP RA with four degrees of freedom of beamforming could be a promising candidate for space and satellite communications.Keywords：Broadband;Dual-band;Dual-circular-polarized;Reflect-array;Shared-aperture
- 4
- |
- 14
- |
- 0
Published：2023-12-01 - Abstract：Under the development of computing network convergence, considering the computing and network resources of multiple providers as a whole in the Computing Force Network (CFN) has gradually become a new trend. However, since each computing and network resource provider (CNRP) only considers its interest and competes with other CNRPs, introducing multiple CNRPs will create the problem of a lack of trust and difficulty in unified scheduling. In addition, concurrent users have different requirements, so there is an urgent need to study how to optimally match users and CNRPs on a many-to-many basis, thus improving user satisfaction and ensuring and improving the utilization of limited resources. In this paper, firstly, we adopt a reputation model based on the beta distribution function to measure the credibility of CNRPs and propose a performance-based reputation update model. Then, we formalize the problem into a constrained multi-objective optimization problem and find the feasible solutions using a modified fast and elitist non-dominated sorting genetic algorithm (NSGA-II). We also conduct extensive simulation experiments to evaluate the proposed algorithm; the simulation results demonstrate that the proposed model and the problem formulation are valid and, based on which NSGA-II algorithm is effective and which can find the Pareto set of CFN, increases user satisfaction and resource utilization. Moreover, a set of solutions provided by the Pareto set give us more choices of the many-to-many matching of users and CNRPs problem according to the actual situation.Keywords：Computing force network;Resource scheduling;Performance-based reputation;User satisfaction
- 17
- |
- 12
- |
- 0
Published：2023-12-01 -
Vibration harmonic suppression technology for electromagnetic vibrators based on an improved sensorless feedback control method Enhanced Publication
Abstract：To realize low harmonic distortion of the vibration waveform output from electromagnetic vibrators, we propose a vibration harmonic suppression technology based on an improved sensorless feedback control method. Without changing the original driving circuit, the AC equivalent resistance of the driving coil is used to obtain high-precision vibration velocity information, and then a simple and reliable velocity feedback control system is established. Through the study of the effect of different values of key parameters on the system, we have achieved an effective expansion of the velocity characteristic frequency band of low-frequency vibration, resulting in an enhanced harmonic suppression capability of velocity feedback control. In this paper we present extensive experiments to prove the effectiveness of the proposed method and make comparisons with conventional control methods. In the frequency range of 0.01 to 1 Hz, without using any sensors, the method proposed in this paper can reduce the harmonic distortion of the vibration waveform by about 40% compared to open-loop control and by about 20% compared to a conventional sensorless feedback control method.Keywords：Vibration calibration;Electromagnetic vibrators;Harmonic suppression;Sensorless control method;Velocity feedback control- 4
- |
- 12
- |
- 0
Published：2023-12-01 - Abstract：Due to factors such as motion blur, video out-of-focus, and occlusion, multi-frame human pose estimation is a challenging task. Exploiting temporal consistency between consecutive frames is an efficient approach to address this issue. Currently, most methods explore temporal consistency through refinements of the final heatmaps. The heatmaps contain the semantic information of key points, which can improve the detection quality to a certain extent. However, they are generated by features, and feature-level refinements are rarely considered. In this paper, we propose a human pose estimation framework with refinements at the feature and semantic levels. We align auxiliary features with the features of the current frame to reduce the loss caused by different feature distributions. An attention mechanism is then used to fuse auxiliary features with current features. In terms of semantics, we use the difference information between adjacent heatmaps as auxiliary features to refine the current heatmaps. The method was validated on the large-scale benchmark datasets PoseTrack2017 and PoseTrack2018, and the results demonstrated the effectiveness of our method.Keywords：Human Pose Estimation;Multi-frame refinement, Heatmap and offset estimation, Feature alignment, Multi-person
- 3
- |
- 15
- |
- 0
Published：2023-12-01 -
Communication Efficiency Optimization of Federated Learning for Computing and Network Convergence of 6G Networks Enhanced Publication
Abstract：Federated learning effectively addresses issues such as data privacy by collaborating across participating devices to train global models. However, factors such as network topology and device computing power can affect its training or communication process in complex network environments. Computing and network convergence (CNC) of sixth generation (6G) networks, a new network architecture and paradigm with computing-measurable, perceptible, distributable, dispatchable, and manageable capabilities, can effectively support federated learning training and improve its communication efficiency. By guiding the participating devices’ training in federated learning based on business requirements, resource load, network conditions, and computing power of devices, CNC can reach this goal. In this article, to improve the communication efficiency of federated learning in complex networks, we study the communication efficiency optimization of federated learning for CNC of 6G networks, methods that gives decisions on its training process for different network conditions and computing power of participating devices. The experiments address two architectures that exist for devices in federated learning and arrange devices to participate in training based on arithmetic power while achieving optimization of communication efficiency in the process of transferring model parameters. The results show that the method we proposed can (1) cope well with complex network situations, (2) effectively balance the delay distribution of participating devices for local training, (3) improve the communication efficiency during the transfer of model parameters, and (4) improve the resource utilization in the network.Keywords：computing and network convergence;communication efficiency;Federated learning;two architectures- 5
- |
- 17
- |
- 0
Published：2023-11-01 - Abstract：In this article, a practical fixed-time adaptive fuzzy control strategy is investigated for uncertain nonlinear systems with time-varying asymmetric constraints and input quantization. To overcome the difficulties of designing controllers under the state constraints case, a unified barrier function approach is employed to construct a coordinate transformation that maps the original constrained system to be an equivalent unconstrained one, thus relaxing the time-varying asymmetric constraints upon system states and avoiding the feasibility check condition typically required in the traditional barrier Lyapunov function-based control approach. Meanwhile, the“explosion of complexity” problem in the traditional backstepping approach arising from repeatedly derivatives of virtual controllers is solved by using the command filter method. It is verified via the fixed-time Lyapunov stability criterion that the system output can track a desired signal within a small error range in a predetermined time, and all system states remain in the constraint range. Finally, a simulation example is offered to demonstrate the effectiveness of the proposed strategy.Keywords：unified barrier function, time-varying asymmetric state constraints, fuzzy logic systems, fixed-time control, command filter
- 4
- |
- 11
- |
- 0
Published：2023-11-01 - Abstract：The synthetic minority oversampling technique (SMOTE) is a popular algorithm to reduce the impact of class imbalance for building classifiers, which has received several enhancements over the past 20 years. SMOTE and its variants synthesize a number of minority-class sample points in the original sample space to alleviate the adverse effects of class imbalance. This approach works well in many cases, but problems arise when synthetic sample points are generated in overlapping areas between different classes, which further complicates classifier training. To address this issue, this paper proposes a novel generalization-oriented rather than imputation-oriented minority-class sample point generation algorithm, named overlapping minimization SMOTE (OM-SMOTE), this algorithm is designed specifically for binary imbalanced classification problems. OM-SMOTE first maps the original sample points into a new sample space by balancing the trade-off between sample encoding and classifier generalization. Then, OM-SMOTE employs a set of sophisticated minority-class sample point imputation rules to generate synthetic sample points that are as far as possible from overlapping areas between classes. Extensive experiments have been conducted on 32 imbalanced data sets to validate the effectiveness of the OM-SMOTE algorithm. Results show that using OM-SMOTE to generate synthetic minority-class sample points leads to better classifier training performances for the naive Bayes classifier, support vector machine, decision tree, and logistic regression than other 11 state-of-the-art SMOTE-based imputation algorithms. This demonstrates that OM-SMOTE is a viable approach to support the training of high-quality classifier for imbalanced classification.Keywords：Imbalanced classification;Synthetic minority oversampling technique;Majority-class sample point;Minority-class sample point;Generalization capability;Overlapping minimization
- 7
- |
- 18
- |
- 0
Published：2023-10-07
- Address：Zhejiang University Press, 148 Tianmushan Road, Hangzhou, China Postal code：310028
- Tel：+86-571-88273162 Email：fitee@zju.edu.cn
- Copyright © 2000 - 2023, Zhejiang University Press 京ICP备09064830号-19
京公网安备11010802024621 - It is recommended to read the content of this site in Chrome&IE9+. Please switch to extreme mode in browser 360.
- Cookies We use cookies to help provide and enhance our service and tailor content. By continuing, you agree to the use of cookies.
0