In the mobile edge computing (MEC)-enabled air-terrestrial integrated network, unmanned aerial vehicles (UAVs) serve as air edge nodes with the purpose to collaboratively train the high-availability prediction model by federated learning (FL). Nonetheless, in view of the significance of data freshness for an accurate training model, UAVs suffer from the stochastic and intermittent nature in energy harvesting (EH). This paper formulates an inter-UAV cooperative federated computing (IUCFC) problem to jointly optimize prediction accuracy, age of information (AoI) in flight region, and overall energy consumption of EH-enabled UAVs for edge data processing. To address the intricate IUCFC problem, a deep reinforcement learning (DRL) based cooperative UAV intelligent decision (CUID) algorithm is proposed, which leverages a dual Actor-Critic architecture, in pursuit of the collective tuning of hybird actions. Further, the Ornstein-Uhlenbeck (OU) noise is engaged in continuous action spaces to prompt exploration, while a conditional iteration dropout (CID) scheme mitigates the infeasible actions caused by the noise introduced, thereby bolstering the exploration efficiency and quality of CUID algorithm. Considering the non-stationary environments originated from UAV mobility, priority experience replay (PER) is adopted to dynamically modify experience priority. Extensive experiments show that CUID attains superior performance over those advanced algorithms, upgrading system utility by 8.79%, while augmenting FL model accuracy by 3.68% in dynamic scenarios with heterogeneous data distributions.

In IoT data sharing, game-theoretic data pricing is widely recognized for its dynamic equilibrium and payoff maximization but faces challenges in real-time performance, integrity, and reliability. This paper proposes a Secure Gametheoretic Framework (SGF) using OP-TEE for secure and efficient game processes, with results uploaded to the blockchain for traceability. The paper also improves the iterative solution of the Stackelberg evolutionary game model to accelerate convergence. Experimental results on the Raspberry Pi 3B shows that the proposed framework completes the pricing process under 3 minutes with approximately 1000 nodes participate in the game, meeting the real-time IoT data sharing requirements. the additional overhead from OP-TEE stays under 15%, balancing security and performance. The improved method reduces time cost by 52.65% and 64.05% compared to existing methods, demonstrating significant effectiveness.

With the increasing attention to privacy protection issues in the field of distributed machine learning, how to empower users with the “right to be forgotten” in federated learning has become an important research direction. We propose a multi feature joint unlearning method to improve privacy protection and model management efficiency in federated learning. By assigning a learnable mask weight to each feature, dynamically adjust the contribution of each feature in the unlearning process. This method not only effectively avoids conflicts between features, but also significantly improves the stability and efficiency of the unlearning process. Meanwhile, sparse regularization ensures the focus of the unlearning process and avoids unnecessary feature unlearning. In addition, the integration of blockchain enhances the transparency and security of the system, ensuring the immutability and traceability of all update operations. In the experimental section, we validated the effectiveness of the method on multiple datasets, particularly on complex datasets where metrics such as accuracy, feature sensitivity, and attack success rate performed well, providing an effective solution for the feature unlearning problem in federated learning.

The identity resolution system serves as the entrance for the Internet of Things (IoT) data, ensuring unique identification of devices and effective data exchange. However, current identity resolution systems face issues such as information silos, uneven distribution of permissions, difficulties in manually managing identifiers for large-scale applications, and security weakness. Therefore, this paper proposes a compatible, fair, automated, and secure identity resolution system named RBBC, which serves as a universal portal for IoT device identification, data retrieval, and data exchanges. RBBC introduces a universal architecture to achieve compatibility in identity resolution, breaking down the information silos within the system; it decentralizes management authority to participants, realizing automated, fair, and equitable identifier management; it utilizes blockchain to ensure the immutability of identifiers and introduces a reputation model with an incentive mechanism into the blockchain to provide a quantifiable and sustainable trust measure for the system, thereby enhancing the security and reliability of identity resolution. Lastly, experimental validation confirms the effectiveness of the reputation model in detecting malicious nodes, and the efficiency of the automatic identifier allocation and registration mechanism, demonstrating the study's suitability for large-scale networks.

The throughput issue has become an urgent problem for large-scale application of blockchain technology. Sharding technology can improve system throughput and scalability. Current research on the optimal blockchain sharding problem uses Model-Free Reinforcement Learning (MFRL) to maximize system throughput. However, these methods learn directly from interactions with the system, resulting in extremely low sampling efficiency. To bridge this gap, this paper presents a Model-based Policy Optimization Blockchain Sharding algorithm (MBPOBS), which utilizes the Gaussian Process Regression (GPR) to precisely predict the future states, based on which the best sharding strategy is learned through imitation learning from the Cross-Entropy Method (CEM) demonstrator. Simulation results show that compared to traditional MRFL algorithms, MBPOBS accelerates the learning by up to 1.8x.

With the continuous expansion of data centers, their carbon emission becomes a serious issue. A number of studies are committing to reduce the carbon emission of data centers. Carbon trading is a promising emission reduction technique which is, however, seldom applied to data centers. To bridge this gap, we propose a carbon-neutral architecture DC2for distributed data centers, where each data center consists of three subsystems, i.e., energy subsystem for energy supply, thermal subsystem for data center cooling and the carbon subsystem for carbon trading. Then, we formulate the energy management problem as a Decentralized Partially Observable Markov Decision Processe (Dec-POMDP) and develop a distributed solution framework using Multi-Agent Deep Deterministic Policy Gradient (MAD-DPG). Finally, simulations on real-world data shows that the DC2provides a cost saving of 8.16%, and applying MADDPG saves 10.04% of the overall cost compared to Independent MADDPG.

In this poster, the energy and carbon management problem in Data Center Microgrid (DCMG) is modeled as a Decentralized Partially Observable Markov Decision Process, and the Multi-Agent Deep Deterministic Policy Gradient (MADDPG) algorithm is adopted to learn the optimal operating policy. In addition, the system uses blockchain and smart contract to ensure data security and auditability in the power trading process. Finally, simulation results show that the proposed system significantly reduces the overall cost of the DCMGs compared with traditional systems.

Mobile Edge Computing (MEC) reduces service latency and enhances Quality of Service (QoS) by offloading tasks to the wireless network edge. However, the rapid growth of task offloading and the associated data transmission security challenges deserve further investigation. This study proposes a blockchain-MEC hybrid solution where mobile devices process tasks and engage in block mining to boost system utility. The objective is to maximize the accumulated reward in the blockchain-MEC system by optimizing off loading decisions, channel selection, transmission power, computing resources, and block intervals. A Markov Decision Process is formulated to model the optimization problem which is solved via a multi-agent deep reinforcement learning (MADRL) algorithm. The results of the simulation demonstrate that our approach is more effective than the baseline method.

Although botnet had been at the top of the list of main threats to the cyber world for an extended period of time, its harmfulness has been constrained nowadays due to the development of kaleidoscopic network security enforcing tools and people’s increasing awareness. And the underlying technology of the botnet has been stagnant ascribing to many drawbacks such as inadequate protection of the identity of the Botmaster and weak resilience of the botnet’s infrastructure. In this article, we first introduce a new classification of the botnet based on botnets’ underlying network, then briefly analyze the main flaws of the traditional botnet and some looming Blockchain-based botnets, with pros and cons of leveraging Blockchain to construct botnets. Furthermore, we propose one IOTA of countless legions (OICL), a newfangled versatile botnet infrastructure that overcomes the bottlenecks that other contemporaries cannot eliminate. It leverages Blockchain, also known as distributed ledger technology (DLT), to be its premises and uses many advantages of it without paying too many tradeoffs. Also, we invent a whole set of communication protocols for OICL and a novel scheme called Proof of Honest (PoH) to identify the espionage infiltrated into the botnet to further promote the robustness. In addition, we discover and propose a mechanism called collateral damage binding (CDB), which proves that the botnet has it such as OICL is far more robust than those who do not. Performance evaluations show that OICL is effective, more cost-saving, and fast-responding compared with the Bitcoin-based botnets as baselines.

With the swift development of express delivery industry, the increasingly attention has been shifted to express delivery mechanism design. Generally, the revenue of the courier is the difference between the users' express fee and the courier's pickup cost. In order to improve the revenue of courier without increasing the user's express fee, this paper presents a low-cost package pickup covering system to find an optimal Hamiltonian pickup tour for the courier over a subset of packages, where packages who are not on the tour should be covered exactly by one package on the tour. A billing rule discounting the express fee to incentivize users to deliver their packages is also proposed. We formulateLow-cost Package Pickup Covering (LPPC)problem to maximize the revenue of the courier. Considering the complexity ofLPPC, we propose aLow-cost Package Pickup Covering Mechanism (LPPCM)to solve theLPPCproblem including problem transformation, hardness analyzing,Attention Model based on Encoder-Decoder Architecture (AMEDA)model design and model training.AMEDAis trained by a deep reinforcement learning algorithm in an unsupervised manner and it can directly output the solution based on the given instances. Through extensive simulations, we demonstrate that the average revenue of courier forAMEDAis at least 10.1% higher than the traditional heuristic local search and is 18.5% lower than the optimal solution on average.AMEDAprovides a desired trade-off between the execution time and solution quality, which is well suited for the large-scale tasks which require quick decisions.

As one of the major carbon producers, data centers produce around 1% of global carbon emissions per year. Researchers are making significant effort to reduce the data center carbon emissions. However, the current carbon-neutral data center solutions seldom take the carbon quotas into account, nor do they integrate new emission reduction technologies like Carbon Capture (CC) and Power-to-Gas (P2G), etc. To bridge this gap, in this paper a novel carbon-neutral data center architecture is proposed based on which a holistic cost minimizing problem is formulated. We use the Variational Mode Decomposition (VMD) and the Long Short-Term Memory (LSTM) neural network to construct highly accurate prediction models, and develop a Model Predictive Control (MPC) algorithm for energy and carbon management. Finally, simulations on real-world data demonstrate that our approach reduces up to 19.19% overall cost compared with traditional solutions.

The rapid advancement in blockchain technology has enabled its applications across wide spectrum of fields. The blockchain throughput, which is usually measured by Transactions Per Second (TPS), is one of the key metrics to reflect the performance of the blockchain systems. However, current blockchain systems have low TPS rates that makes them unsuitable for latency critical applications like Vehicle-to-vehicle (V2V) communication. To address the above issue, the sharding technology, which divides the network into multiple disjoint groups so that transactions can be processed in parallel, is applied to the blockchain systems as a promising solution to improve TPS. This paper considers the Optimal Blockchain Sharding (OBCS) problem which is formulated as a Markov Decision Process (MDP) where the decision variables are the number of shards, block size and block interval. Previous works solved the OBCS problem via Deep Reinforcement Learning (DRL) based methods where the action space has to be discretized such that it is not too large for tractability. However, the discretization degrades the solution quality since the optimal solution usually lies between discrete values. In this paper, we treat the block size and block interval as continuous decision variables and propose a sharding control algorithm based on Parametrized Deep Q-Networks (P-DQN) to efficiently handle the discrete-continuous hybrid action space without the scalability issue. Experimental results show that our Parametrized Deep Q-Networks Blockchain Sharding (P-DQNBS) method can effectively improve the TPS by up to 20%.

The growth in scale and power density of Data Centers (DC) poses serious challenges to the cooling management. Recently, there are many studies using machine learning to solve the cooling management problems. However, a comprehensive comparative study is still missing. In this work, we compare the performance of various Deep Reinforcement Learning (DRL) algorithms, including Deep-Q Networks (DQN), Deep Deterministic Policy Gradient (DDPG), and Branching Dueling Q-Network (BDQ), using the Active Ventilation Tiles (AVTs) control problem in raised-floor DC as an example. In particular, we design two multiagent algorithms based on DQN and three critic architectures for DDPG. Simulations based on real world workload show that DDPG provides the best performance over the considered algorithms.