Training machine learning (ML) models for scientific problems is often challenging due to limited observation data. To overcome this challenge, prior works commonly pre-train ML models using simulated data before having them fine-tuned with small real data. Despite the promise shown in initial research across different domains, these methods cannot ensure improved performance after fine-tuning because (i) they are not designed for extracting generalizable physics-aware features during pre-training, (ii) the features learned from pre-training can be distorted by the fine-tuning process. In this paper, we propose a new learning method for extracting, preserving, and adapting physics-aware features. We build a knowledge-guided neural network (KGNN) model based on known dependencies amongst physical variables, which facilitate extracting physics-aware feature representation from simulated data. Then we fine-tune this model by alternately updating the encoder and decoder of the KGNN model to enhance the prediction while preserving the physics-aware features learned through pre-training. We further propose to adapt the model to new testing scenarios via a teacher-student learning framework based on the model uncertainty. The results demonstrate that the proposed method outperforms many baselines by a good margin, even using sparse training data or under out-of-sample testing scenarios.

Federated learning, leveraging distributed data from multiple nodes to train a common model, allows for the use of more data to improve the model while also protecting the privacy of original data. However, challenges still exist in ensuring privacy and security within the interactions. To address these issues, this paper proposes a federated learning approach that incorporates blockchain, homomorphic encryption, and reputation. Using homomorphic encryption, edge nodes possessing local data can complete the training of ciphertext models, with their contributions to the aggregation being evaluated by a reputation mechanism. Both models and reputations are documented and verified on the blockchain through consensus process, which then determines the rewards based on the incentive mechanism. This approach not only incentivizes participation in training, but also ensures the privacy of data and models through encryption. Additionally, it addresses security risks associated with both data and network attacks, ultimately leading to a highly accurate trained model. To enhance the efficiency of learning and the performance of the model, a joint adaptive aggregation and resource optimization algorithm is introduced. Finally, simulations and analyses demonstrate that the proposed scheme enhances learning accuracy while maintaining privacy and security.

The increasingly intense cyber-attacks have always been a crucial issue to the communication-based train control (CBTC) system due to exposed wireless channels. Both cyber-attack intrusion detection and defense policy calculation demand substantial computing resources. Combined with high capacity and reliability 5G technologies, edge intelligence (EI) is believed to help empower CBTC systems in terms of security and efficiency. This paper proposes an EI-enabled structure for CBTCs to defend against cyber-attacks, where the EI server provides real-time intelligent computing services for trains to derive real-time defense policies. We formulate the cyber-attack and defense process in EI-enabled CBTCs as a two-level game model, where system security and edge computing resource allocation are jointly optimized. In the lower-level game, we model interactions between the cyber attacker and system defender as a discrete repeated security game (DRSG), which is also a non-zero sum and incomplete information game. The fictitious play (FP) is introduced to derive a Nash equilibrium (NE) based optimal defense scheme. In the upper-level game, considering that the EI server cannot simultaneously update the optimal defense scheme for all trains due to the limited computation resources, we construct a multi-stage computation resource allocation game (MCRAG). We derive the optimal computation resource allocation scheme by the neural fictitious self-play (NFSP), where a deep Q-learning network (DQN) and a supervised learning network are jointly built to learn the strategy. Extensive simulation results show that our proposed EI-enabled CBTC system and the two-level game model can effectively defend against various attacks.

Autonomous driving is a core application that greatly benefits from Internet of Vehicles (IoV). The calculation of the precise positions of Connected Autonomous Vehicles (CAVs) is mainly done using a Deep Neural Network (DNN) which requires significant computing power. Therefore, reducing the computational overhead and improving the efficiency are urgent problems to be solved. In this paper, we first propose a CAV cooperative learning architecture based on blockchain to improve the positioning accuracy of vehicles. Then, we introduce an error precision sharing model between CAVs. The proposed framework enables CAVs to train vehicle positioning accuracy models locally and exchange them via a blockchain network. Such a distributed training architecture further reduces the computing power required. Extensive simulation results show that the proposed scheme can also significantly improve the accuracy of the trajectory error compared to existing approaches.

Recently, Internet of Vehicles (IoV) and Machine Learning (ML) have attracted more and more attention. Considering inefficient real-time training and high requirements on computing capabilities of centralized data collection, performing Distributed Machine Learning (DML) in IoV has become an important research branch. However, the heterogeneity, mobility, and distrust among IoV nodes affect how to execute DML effectively, securely, and in a salable manner. In this paper, a blockchain-based Cooperative Learning framework combined with a Deep Compression method (CLDC) is proposed. First, we improve the local training efficiency of lightweight IoV nodes by using deep compression method. Meanwhile, we have introduced a blockchain system in CLDC, the significance of which is that we have completed the transformation from centralized architecture to distributed framework through the blockchain, and shared local training results in a verifiable manner. The framework uses non-tamperable features of the blockchain to ensure the security of local training results. Moreover, we propose a Learning-based Redundant Byzantine Fault Tolerance (L-RBFT) protocol, in which the primary node needs to confirm the loss percentage of learning in the transaction before forwarding the RBFT messages. The significance of L-RBFT is to ensure that IoV nodes obtain the best training results through the consensus of blockchain nodes. We use it to solve the computing and communication resource allocation problem in IoV to clarify the operating mechanism of the proposed framework. The experimental results prove that this scheme performs better when compared with the traditional centralized deep reinforcement learning method.

Vehicular ad hoc networks (VANETs) have become an important branch of future 6G smart wireless communications. As an emerging key technology, multi-access edge computing (MEC) provides low-latency, high-speed, and high-capacity network services for the VANETs. In this paper, we propose a novel framework for blockchain-based, hierarchical multi-access edge computing for the future VANET ecosystem (BMEC-FV). In the underlying VANET environment, we propose a trust model to ensure the security of the communication link between vehicles. Multiple MEC servers calculate the trust between vehicles through computing offloading. Meanwhile, the blockchain system plays an important role to manage the entire BMEC-FV architecture. We aim to optimize the throughput and the quality of services (QoS) for MEC users in the lower layer of the system architecture. In this framework, the main challenge is how to effectively reach consensus among blockchain nodes while ensuring the performance of MEC systems and blockchains. The blocksize of blockchain nodes, the number of consensus nodes, reliable features of each vehicle, and the number of producing blocks for each block producer are considered in a joint optimization problem, which is modeled as a Markov decision process with state space, action space, and reward function. Since it is difficult for this to be solved by traditional methods, we propose a novel deep compressed neural network scheme. Simulation results illustrate the superiority of the BMEC-FV ecosystem.

The appropriate design of a vehicular ad hoc network (VANET) has become a pivotal way to build an efficient smart transportation system, which enables various applications associated with traffic safety and highly-efficient transportation. VANETs are vulnerable to the threat of malicious nodes stemming from its dynamicity and infrastructure-less nature and causing performance degradation. Recently, software-defined networking (SDN) has provided a feasible way to manage VANETs dynamically. In this article, we propose a novel software-defined trust based VANET architecture (SD-TDQL) in which the centralized SDN controller is served as a learning agent to get the optimal communication link policy using a deep$Q$-learning approach. The trust of each vehicle and the reverse delivery ratio are considered in a joint optimization problem, which is modeled as a Markov decision process with state space, action space, and reward function. Specifically, we use the expected transmission count ($ETX$) as a metric to evaluate the quality of the communication link for the connected vehicles’ communication. Moreover, we design a trust model to avoid the bad influence of malicious vehicles. Simulation results prove that the proposed SD-TDQL framework enhances the link quality.

Vehicular ad hoc networks (VANETs) have become a promising technology in smart transportation systems with rising interest of expedient, safe, and high- efficient transportation. Dynamicity and infrastructure-less of VANETs make it vulnerable to malicious nodes and result in performance degradation. In this paper, we propose a software- defined trust based deep reinforcement learning framework (TDRL-RP), deploying a deep Q-learning algorithm into a logically centralized controller of software-defined networking (SDN). Specifically, the SDN controller is used as an agent to learn the highest routing path trust value of a VANET environment by convolution neural network, where the trust model is designed to evaluate neighbors' behaviour of forwarding packets. Simulation results are presented to show the effectiveness of the proposed TDRL-RP framework.