The integration of Domain Knowledge Graphs (DKG) into Retrieval-Augmented Generation has emerged as a promising approach for constructing Specialized Large Language Models (spLLMs).On account of the scarcity of high-quality DKGs, existing approaches employ an evolutionary framework, wherein the DKG is continuously evolved alongside its utilization for enhancing the LLM. Yet, these methods face two key limitations: 1) heavy reliance on costly expert knowledge, and 2) neglect of the connection between the LLM's inherent knowledge and external expert knowledge. To address these issues, this paper introduces Epistemic Cognition-enhanced Specialized LLMs (EC-spLLM), a novel evolutionary framework that instills epistemic cognition into the LLM to systematically exploit both its internal knowledge and expert knowledge. At the core of EC-spLLM lies the Hallucination Detection-based Epistemic Cognition (HDEC) mechanism, which assesses the reliability of LLM-generated responses using the LLM's self cognition and hallucination detection. This assessment ability enables EC-spLLM to selectively adopt either the expert-provided golden answer or a reliable LLM-generated answer during DKG evolution, thereby reducing dependence on experts and bridging internal and external knowledge sources to enhance performance. We conducted extensive experiments on five datasets spanning five domain, e.g., emotional sociology, biology, ect. Results show that EC-spLLM reduces the usage of golden answers by an average of 67% while retaining 97.2% of the accuracy achieved by the SOTA method, and outperforms all other baselines.

Prompting collaboration is a promising choice to enhance the performance of Vehicle Trajectory Prediction (VTP) in the Internet-of-Vehicles (IoV). This paper designs an edge-assisted collaborative VTP framework that employs Roadside Units (RSU) to associate vehicles and encourages RSUs to collaborate to realize VTP. A novel decentralized machine learning method, called Hierarchical Swarm Learning (HierSL), is proposed to improve efficiency and security in the collaborative learning process. HierSL is particularly suited for large-scale edge-assisted IoV systems with its two-layer hierarchical learning framework. HierSL allows nearby RSUs to form local organizations and performs lower-layer learning (local level) for knowledge integration within each organization. Over local organizations, an upper layer is constructed to integrate the knowledge of local organizations, thereby generating well-performed global models. Compared to vanilla Swarm Learning (SL), HierSL not only reduces the reliance on global communications but also cuts the cost of blockchain for collaborative VTP. Experiments are conducted on a real-world NGSIM US-101 data set, and the results show that the proposed method outperforms vanilla SL and is comparable to centralized learning.

In-network content caching is a fundamental functionality in Information Centric Network (ICN), which facilitates content distribution with reduced bandwidth consumption, lower network congestion, and faster content retrieval. However, traditional heuristic caching strategies often fail to handle dynamic ICN environments, and most leaning-based strategies cannot realize secured content caching for distributed ICN. In observation of these challenges, this paper investigates collaborative content caching in ICN, and proposes a novel algorithm, called Swarm Reinforcement Learning (SRL), for designing a secured caching mechanism in distributed ICN caching platform. SRL inherits salients features from both Swarm Learning (SL) and Deep Reinforcement Learning (DRL): it enables a fully decentralized learning process and strictly guarantees the privacy and security of data and models during collaborative caching by leveraging the blockchain technique; SRL also lets local ICN router interact with the local environment and integrate the learned knowledge with other ICN routers for constructing a collaborative caching strategy that maximizes the long-term reward of the entire ICN caching platform. We carry out systematic experiments to evaluate the performance of the proposed method. The results show that SRL-based collaborative caching outperforms state-of-the-art caching strategies in terms of cache hit rate and content retrieval delay, and also improves the stability of the ICN caching platform.

Adaptive traffic control systems are gaining attention in recent years as traditional hand-crafted traffic control experiences performance fall-offs with increasingly complicated metropolitan traffic patterns. This paper studies a learning automata (LA)-based traffic signal control scheme that adapts to real-time traffic patterns and optimizes traffic flows by dynamically changing the green split timings. A novel LA algorithm, called K-Neighbor Multi-Agent Learning Automata (KN-MALA), is proposed to learn the optimal decision online and adjust the traffic light accordingly in an attempt to minimize the overall waiting time at an intersection. In particular, KN-MALA employs an online distributed learning framework that integrates the traffic condition of neighboring intersections to efficiently learn and infer optimal decisions for large-scale traffic signal systems. Furthermore, a parameter insensitive update mechanism is designed for KN-MALA to overcome the instability caused by initialization variations. Experiments are conducted on real-world traffic patterns of Sioux Falls City and the performance of the proposed algorithm is compared with the pre-timed traffic light control scheme and an adaptive traffic light control scheme based on single-agent learning automata. The results show that the proposed algorithm outperforms the other schemes in terms of quick traffic clearance under various traffic patterns and initial conditions.

Intrusion detection can distinguish between normal traffic and intrusions. It can be abstracted as a classification problem to use deep learning tools. However, there are unknown intrusions in the real network environment, and their detection is an open set recognition problem. Existing methods such as OpenMax only calculate classification probabilities more smartly, but they can’t explicitly teach the network to recognize features of unknown classes. To improve the ability of deep learning tools to deal with intrusion detection problems based on the open set, the KCC (Known Central Clustering) method is proposed in this paper and is compatible with the existing network architectures. Considering that the high-dimensional vectors in the embedding space represent the deep features, different classes of intrusions in this space have their clusters. By introducing CD-loss (Class Distance-loss), we can get the centers of different class clusters. By introducing negative samples as the unknown classes for training, we can get the threshold of the known classes and reject unknown intrusions by comparing them with fuzzy distance. Experiments on CIC-IDS2017 and CIC-DDoS2019 datasets show that the KCC method improves the classification accuracy of known intrusions and reduces the misclassification rate of unknown intrusions.

Heterogeneous graph embedding is to learn node representations that preserve both heterogeneity and semantic information of a heterogeneous graph, so as to facilitate various downstream tasks, such as node classification and link prediction. Most of the existing methods process the graph structure based on the meta-path to learn low-dimensional feature vectors. However, these methods need to manually set the meta-path which requires strong prior knowledge and omit the content features of the intermediate nodes along the meta-path. In this paper, we propose a novel heterogeneous graph neural network based on node position information (PHGNN). With this method, the meta-path is no longer required, and heterogeneous information can be automatically preserved. Specifically, the proposed framework consists of two main components: content-level learning and type-level learning. Content-level learning utilizes the features information of nodes to learn the importance of the edge-type based nodes to the target node, while type-level learning focuses on the importance of different types of nodes. With the two-level mechanism, PHGNN can hierarchically aggregate multi-type information. The experiments on multiple public datasets show that compared with state-of-the-art baselines, our proposed PHGNN model achieves significant and consistent improvements.

Community detection is an important technique to understand the structure of complex social networks. Many approaches have been devised to extract community structures in recent years. In this paper we propose a novel neighborhood vector propagation algorithm (NVPA) to detect communities in a social network which has greater accuracy than algorithms in the literature. In our approach, a neighborhood vector is proposed to store the neighborhood information, and a vector propagation algorithm is designed to disseminate neighborhood information to other nodes. After neighborhood propagation, hierarchical clustering is used to find the community structure based on similarity measures. We apply our algorithm on two real-world networks and LRF benchmark networks. Experimental results show that our algorithm achieves greater accuracy than several well known algorithms in the literature.

In a communication system with time varying channel qualities, it is ideal to allocate the limited transmission power to channels that will be in good state. However, it is very challenging to do so because channel states are usually unknown when the power allocation decision is made. In this paper, we derive an optimal power allocation policy that can maximize the expected discounted number of bits transmitted over an infinite time span. Specifically, we first model this problem as a partially observable Markov decision processes (POMDP), and analytically investigate the structure of the optimal policy. Then a simple threshold-based policy is derived for a three-channel communication system. By formulating and solving a linear programming formulation of this power allocation problem, we further verified the derived structure of the optimal policy.

We study the fundamental problem of optimal power allocation over two identical Gilbert-Elliott (Binary Markov) communication channels. Our goal is to maximize the expected discounted number of bits transmitted over an infinite time span by judiciously choosing one of the four actions for each time slot: 1) allocating power equally to both channels, 2) allocating all the power to channel 1, 3) allocating all the power to channel 2, and 4) allocating no power to any of the channels. As the channel state is unknown when power allocation decision is made, we model this problem as a partially observable Markov decision process(POMDP), and derive the optimal policy which gives the optimal action to take under different possible channel states. Two different structures of the optimal policy are derived analytically and verified by linear programming simulation. We also illustrate how to construct the optimal policy by the combination of threshold calculation and linear programming simulation once system parameters are known.

We study the problem of power allocation over two identical Gilbert-Elliot communication channels. Our goal is to maximize the expected discounted number of bits transmitted over an infinite time horizon. This is achieved by choosing among three possible strategies: (1) betting on channel 1 by allocating all the power to this channel, which results in high data rate if channel 1 happens to be in good state, and zero bits transmitted if channel 1 is in bad state (even if channel 2 is in good state) (2) betting on channel 2 by allocating all the power to the second channel, and (3) a balanced strategy whereby each channel is allocated half the total power, with the effect that each channel can transmit a low data rate if it is in good state. We assume that each channel's state is only revealed upon transmission of data on that channel. We model this problem as a partially observable Markov decision processes (MDP), and derive key threshold properties of the optimal policy. Further, we show that by formulating and solving a relevant linear program the thresholds can be determined numerically when system parameters are known.

MIMO (Multiple-Input-Multiple-Output) is a promising structure for wireless communication. While virtual MIMO structure has been proposed for distributed and cooperative wireless networks, this proposed structure has also put additional energy consumption where energy stands in the important position. At the same time, we have observed that this-problem-caused low energy nodes' mibehavior will degrade the whole system efficiency. In this paper, we propose a two-tier correlation matrix based low power nodes detection system which can capture and mitigate relays' malicious behavior before signal combining. This mechanism can effectively find out the low power node in virtual MIMO structure and thus improve the system efficiency. The simulation results show that a better bit error rate performance of this structure in the presence of low power node and our detection mechanism as well is achieved.

The theory of authentication tests is a powerful tool for analyzing and designing cryptographic protocols. However, it is difficult to apply the theory directly to prove the security goals of the protocols because determining the type of the test (e.g. outgoing, incoming and unsolicited test) is a little complex for computer and deriving the security properties of a test needs more intelligence. Therefore, automatic security protocol analyzer cannot be implemented efficiently based on this theory. To solve this problem, in this paper we propose an authentication test type matrix (ATTM), which makes the identification of the test type very simple and straightforward. Furthermore, we propose and prove a set of security properties associated with each case indicated by the elements in the ATTM, which can be used directly in protocol analysis and design. Using the example of Needham-Schroeder protocol, we demonstrate that ATTM makes automatic security protocol verification and design much easier and more straightforward.

In this paper, we present an anonymity scheme to achieve all the peers' anonymity in P2P reputation systems by changing pseudonym with the help of a trusted third party (TTP) server, such that neither the TTP server nor other peers can link each peer's pseudonyms. Compared with RuP, an existing representative scheme, our scheme has two main features. First, the TTP server's cost is reduced by more than half in terms of numbers of RSA decryption operations and message overhead, which makes our scheme be more scalable. Second, we also propose a reputation metric that can be integrated into P2P reputation systems that achieve peers' anonymity by changing pseudonym, to help estimate the trustworthiness of peers. Simulation results show the effectiveness of our proposed reputation metric.

Much work has been done to modify the TCP protocol to improve TCP performance - mainly throughput, over wireless link. However, such improvements need to change the TCP implementation on user systems, which is usually difficult to deploy. In this paper, we investigate the effectiveness of enhancing the TCP performance over wireless link by tuning the packet scheduling schemes. Based on an opportunistic packet scheduling algorithm we proposed, we demonstrate that TCP throughput can be significantly improved by using channel-aware packet scheduling algorithms without changing the TCP implementation. Our work suggests that in cases where it is impractical to change the TCP protocols, packet scheduling is an effective way to improve the TCP performance over wireless links.

In this paper, we propose a trust model for the open Grid market. Our trust model emphasizes the importance of both direct trust and indirect trust/reputation when evaluating the trustworthiness of a Grid service provider. Since many factors contribute to the direct and indirect trust, we propose a novel method based on multiple attribute decision making (MADM) theory to determine the objective weights of both direct and indirect trust. The simulation results demonstrate that our trust model reflects the trustworthiness of a service provider more accurately than the weighted feedback model and eBay trust model, and thus improves user satisfaction in the open Grid environment.

This paper presents a novel video scheduling algorithm for streaming MPEG video over multi-user shared wireless link, in which video sessions are scheduled based on both location-dependent channel condition and instantaneous frame delay. This algorithm also employs a pro-active frame discard mechanism to discard potential problematic frames. The numerical results show that the proposed algorithm is able to significantly improve link throughput and guarantee long-term fairness.