The emergence of text-to-image models has recently sparked significant interest, but the attendant is a looming shadow of potential infringement by violating user terms. Specifically, an adversary may exploit data created by a commercial model to train their own without proper authorization. To address such risk, it is crucial to investigate the attribution of a suspicious model's training data by determining whether its training data originates, wholly or partially, from a specific source model. To trace the generated data, existing methods need to apply additional watermarks during either the training or inference phases of the source model. However, these methods are impractical for pre-trained models that have been released, especially when model owners lack security expertise. To tackle this challenge, we propose an injection-free training data attribution method for text-to-image models. It can identify whether a model's training data stems from a certain source model without adding additional watermarks on the source model. The rationale of our method lies in the inherent memorization characteristic of text-to-image models. The memorization of training data is inherited through the data generated by the source model to the model trained on that data, making the source model and the infringing model exhibit consistent behaviors on specific samples. Therefore, from instance-level, we develop detection-based and generation-based strategies to uncover these distinct samples and using them as inherent watermarks to verify if a suspicious model originates from the source model. Besides, we also propose a statistical-level attribution method, utilizing the shadow model technique to train an attribution discriminator. Experiments demonstrate that the attribution accuracy and AUC scores of our methods are over 80% even when the infringing model only uses a small proportion of generated data.

In the Internet of Vehicles (IoV), developing accurate road information models is essential for analyzing perception data gathered from multiple vehicles. However, traditional centralized data-sharing methods can compromise the privacy and security of data providers. federated learning (FL) presents a promising solution as a distributed machine learning approach that balances data privacy protection with efficient utilization by keeping data localized and sharing only model updates. Nevertheless, conventional FL strategies often fail to adequately address differences in resource investment and data quality among participating vehicles while aggregating local training results. This oversight can lead to inequitable model aggregation and distribution, reducing the motivation for vehicles to share their data. This article proposes a reputation evaluation-based, fair, and secure FL scheme for the IoV to address these challenges. In this scheme, the aggregation node utilizes fuzzy comprehensive evaluation to assess the training outcomes of participating vehicles and assigns aggregation weights accordingly. It also calculates reputation values for each vehicle using periodic averaging methods. Subsequently, the node implements differentiated global model compression and distribution based on these reputation scores. Experimental results indicate that the proposed scheme performs comparably to established algorithms while effectively evaluating vehicle reputations. It achieves model compression and equitable distribution, demonstrating an ability to identify and counteract malicious client attacks. Consequently, this approach enhances fairness and security in FL systems designed for the IoV.

To realize data sovereignty, the International Data Space (IDS), adopting usage policies to determine how, when and where other enterprises or individuals may use data, has been proposed by the IDS association and widely received attention from academia and industry. However, because data in the IDS are transferred across domains, existing policy creation approaches for a single domain cannot be applied in the IDS. To address this problem, in this paper, we propose a negotiation scheme to create usage policies in the IDS. In detail, we formulate usage policy negotiation as a combinatorial auction problem and adopt the Vickrey-Clarke-Groves (VCG) mechanism to incentivize potential data providers to truthfully negotiate usage policies. Both theoretical and simulation results show that our scheme maintains truthfulness on data providers and is cost-efficient.

Network functionality is growing increasingly complex, making the commands addition of network device a steadily growing challenge. In order to improve the efficiency of adding new commands to developing or developed network devices, we propose an extensible command parsing method (ECPM). With this method, we design the extensible command parsing system consisting of business-logic module and command-matching module. Business-logic module includes user-defined command rule files and a rule file parsing program. The grammar of user-defined command rule is concise and easy. The rule file parsing program parses rule files into command tree which is disposed in the memory. Command-matching module consists of user commands handling program and command tokens matching program. The former program divides user commands into command tokens. The latter program matches command tokens with command tree to locate the command handling function. We present the design of such a parser using ECPM and apply it to new network devices. The result shows that the ECPM improves the variability, extensibility and flexibility of network devices parsing system.

With the wide application of the network and the rapid expansion of its scale, the types of network devices deployed in one network are getting more and more as well. Each kind of devices has its own management method, so there is an urgent need for a unified network device management system. This paper designs an extensible network device management system, which is able to manage multi-type network devices and is easy to extend its functions. The system achieves these features by loading corresponding components dynamically. It provides a component interface specification, and components developed according to the specification can be loaded to the system flexibly. The system also defines a communication protocol for admission control and management data transmission. The protocol has an advantage in security and extensibility. Experiments show that the system can manage common network devices properly and flexibly. It also shows that the system improves the convenience and security of the network device management.