As large language models (LLMs) are increasingly deployed on resource-constrained edge devices, quantization techniques have been widely adopted to reduce model size and computational requirements. However, this process can expose models to new vulnerabilities. In this work, we introduce the Quantization Conditioned Misalignment (Q-Misalign) attack, a novel threat in which safety misalignment remains dormant in a full-precision LLM but becomes exploitable post-quantization. We demonstrate that our Q-Misalign attack effectively bypasses safety mechanisms and enables the generation of harmful content in quantized models while maintaining full-precision performance. Furthermore, we propose a contrastive task vector-based approach to enhance attack durability, ensuring that vulnerabilities persist even after downstream fine-tuning. Experimental results show that Q-Misalign attack significantly increases jailbreak success rates in quantized models, while preserving model utility and safety alignment in full precision. Our findings highlight a critical gap in current LLM safety measures and call for more robust defenses in quantization-aware scenarios.

Incorporating user preferences into large language models (LLMs) can enhance the personalization and reliability of model outputs and facilitate the application of LLMs to real-world scenarios. However, leveraging user preferences can be a double-edged sword. Recent studies have found that improper utilization can incur sycophancy, where LLMs prioritize alignment with user preferences over the correctness of their outputs. To address sycophancy in LLMs, we analyze and model the problem through the lens of structured causal models (SCMs). We attribute sycophancy to LLMs' reliance on spurious correlations between user preferences and model outputs in this paper. Based on the proposed SCMs, we develop a novel framework, termed **CAUSM**, to mitigate sycophancy in LLMs by exploiting a significant causal signature. Specifically, we eliminate the spurious correlations embedded in the intermediate layers of LLMs through causally motivated head reweighting, and then calibrate the intra-head knowledge along the causal representation direction. Extensive experiments are conducted across diverse language tasks to demonstrate the superiority of our method over state-of-the-art competitors in mitigating sycophancy in LLMs.

Recent diffusion models have manifested extraordinary capabilities in generating high-quality, diverse, and innovative images guided by textual prompts. Nevertheless, these state-of-the-art models may encounter the challenge of concept bleeding when generating images with multiple entities or attributes in the prompt, leading to the unanticipated merging or overlapping of distinct objects in the synthesized result. The current work exploits auxiliary networks to produce mask-constrained regions for entities, necessitating the training of an object detection network. In this paper, we investigate the bleeding reason and find that the cross-attention map associated with a specific entity or attribute tends to extend beyond its intended focus, encompassing the background or other unrelated objects and thereby acting as the primary source of concept bleeding. Motivated by this, we propose Entity Localization and Anchoring (ELA) to drive the entity to concentrate on the expected region accurately during inference, eliminating the necessity for training. Specifically, we initially identify the region corresponding to each entity and subsequently employ a tailored loss function to anchor entities within their designated positioning areas. Extensive experiments demonstrate its superior capability in precisely generating multiple objects as specified in the textual prompts.

Recent advancements in vision-language-to-image (VL2I) diffusion generation have made significant progress. While generating images from broad vision-language inputs holds promise, it also raises concerns about potential misuse, such as copying artistic styles without permission, which could have legal and social consequences. Therefore, it's crucial to establish governance frameworks to ensure ethical and copyright integrity, especially with widely used diffusion models. To address these issues, researchers have explored various approaches, such as dataset filtering, adversarial perturbations, machine unlearning, and inference-time refusals. However, these methods often lack either scalability or effectiveness. In response, we propose a new framework called causal representation editing (CRE), which extends representation editing from large language models (LLMs) to diffusion-based models. CRE enhances the efficiency and flexibility of safe content generation by intervening at diffusion timesteps causally linked to unsafe concepts. This allows for precise removal of harmful content while preserving acceptable content quality, demonstrating superior effectiveness, precision and scalability compared to existing methods. CRE can handle complex scenarios, including incomplete or blurred representations of unsafe concepts, offering a promising solution to challenges in managing harmful content generation in diffusion-based models.

Federated Knowledge Graph Embedding (FKGE) is an emerging collaborative learning technique for deriving expressive representations (i.e., embeddings) from client-maintained distributed knowledge graphs (KGs). However, poisoning attacks in FKGE, which lead to biased decisions by downstream applications, remain unexplored. This paper is the first work to systematize the risks of FKGE poisoning attacks, from which we develop a novel framework for poisoning attacks that force the victim client to predict specific false facts. Unlike centralized KGEs, FKGE maintains KGs locally, making direct injection of poisoned data challenging. Instead, attackers must create poisoned data without access to the victim's KG and inject it indirectly through FKGE aggregation. Specifically, to create poisoned data, the attacker first infers the targeted relations in the victim's local KG via a new KG component inference attack. Then, to accurately mislead the victim's embeddings via aggregation, the attacker locally trains a shadow model using the poisoned data and uses an optimized dynamic poisoning scheme to adjust the model and generate progressive poisoned updates. Our experimental results demonstrate the attack's effectiveness, achieving a remarkable success rate on various KGE models (e.g., 100% on TransE with WN18RR) while keeping the original task's performance nearly unchanged.

The recent success of pre-trained language models (PLMs) such as BERT has resulted in the development of various beneficial database middlewares, including natural language query interfaces and entity matching. This shift has been greatly facilitated by the extensive external knowledge of PLMs. However, as PLMs are often provided by untrusted third parties, their lack of standardization and regulation poses significant security risks that have yet to be fully explored. This paper investigates the security threats posed by malicious PLMs to these emerging database middleware. We specifically propose a novel type of Trojan attack, where a maliciously designed PLM causes unexpected behavior in the database middleware. These Trojan attacks possess the following characteristics: (1) Triggerability: The Trojan-infected database middleware will function normally with normal input, but will likely malfunction when triggered by the attacker. (2) Imperceptibility: There is no need for noticeable modification of the input to trigger the Trojan. (3) Generalizability: The Trojan is capable of targeting a variety of downstream tasks, not just one specific task. We thoroughly evaluate the impact of these Trojan attacks through experiments and analyze potential countermeasures and their limitations. Our findings could aid in the creation of stronger mechanisms for the implementation of PLMs in database middleware.

Generative models produce unbounded outputs, necessitating the use of refusal techniques to confine their output space. Employing generative refusals is crucial in upholding the ethical and copyright integrity of synthesized content, particularly when working with widely adopted diffusion models. "Concept negation'' presents a promising paradigm to achieve generative refusals, as it effectively defines and governs the model's output space based on concepts, utilizing natural language interfaces that are readily comprehensible to humans. However, despite the valuable contributions of prior research to the field of concept negation, it still suffers from significant limitations. The existing concept negation methods, which operate based on the composition of score or noise predictions from the diffusion process, are limited to independent concepts (e.g., ``a blonde girl`` without ``glasses``) and fail to consider the interconnected nature of concepts in reality (e.g., ``Mickey mouse eats ice cream`` without ``Disney characters``). Keeping the limitations in mind, we propose a novel framework, called $ProtoRe$, to improve the flexibility of concept negation via test-time negative concept identification along with purification in the feature space. $ProtoRe$ works by incorporating CLIP's language-contrastive knowledge to identify the prototype of negative concepts, extract the negative features from outputs using the prototype as a prompt, and further refine the attention maps by retrieving negative features. Our evaluation on multiple benchmarks shows that $ProtoRe$ outperforms state-of-the-art methods under various settings, in terms of the effectiveness of purification and the fidelity of generative images.

The paradigm of Internet of Things (IoT) has attracted much attention in both academia and industry areas in the past decades. Recently, the integration of IoT and social networks is advocated to promote further development. This article focuses on the applications, solutions, and challenges of Social Internet of Things (SIoT) over collaborative edge computing, which exploits the advantages of both mobile edge computing and social relationships among SIoT users. First, applications of collaborative offloading, caching, and streaming data processing are illustrated, followed by representative social-aware solutions, including auction, coalition game, and federated learning. Finally, several research challenges are described. The main contributions of this article are as follows: 1) we specify the role of social ties in traditional applications of IoT and its impacts on individual’s selection; 2) we elaborate the reasons why the presented three approaches can be applied in SIoT; and 3) the discussed challenges can contribute to the future development of secure, robust, and intelligent SIoT frameworks.

A variety of novel mobile applications are developed to attract the interests of potential users in the emerging 5G-enabled vehicular networks. Although computation offloading and task scheduling have been widely investigated, it is rather challenging to decide the optimal offloading ratio and perform adaptive task scheduling in high-dynamic networks. Furthermore, the scheduling policy made by the network operator may be violated, since vehicular users are rational and selfish to maximize their own profits. By considering the incentive compatibility and individual rationality of vehicular users, we present POETS, an efficient partial computation offloading and adaptive task scheduling algorithm to maximize the overall system-wide profit. Specially, a two-sided matching algorithm is first proposed to derive the optimal transmission scheduling discipline. After that, the offloading ratio of vehicular users can be obtained through convex optimization, without any information of other users. Furthermore, a non-cooperative game is constructed to derive the payoff of vehicular users that can reach the equilibrium between users and the network operator. Theoretical analyses and performance evaluations based on real-world traces of taxies demonstrate the effectiveness of our proposed solution.

The prompt evolution of Internet of Medical Things (IoMT) promotes pervasive in-home health monitoring networks. However, excessive requirements of patients result in insufficient spectrum resources and communication overload. Mobile Edge Computing (MEC) enabled 5G health monitoring is conceived as a favorable paradigm to tackle such an obstacle. In this paper, we construct a cost-efficient in-home health monitoring system for IoMT by dividing it into two sub-networks, i.e., intra-Wireless Body Area Networks (WBANs) and beyond-WBANs. Highlighting the characteristics of IoMT, the cost of patients depends on medical criticality, Age of Information (AoI) and energy consumption. For intra-WBANs, a cooperative game is formulated to allocate the wireless channel resources. While for beyond-WBANs, considering the individual rationality and potential selfishness, a decentralized non-cooperative game is proposed to minimize the system-wide cost in IoMT. We prove that the proposed algorithm can reach a Nash equilibrium. In addition, the upper bound of the algorithm time complexity and the number of patients benefiting from MEC is theoretically derived. Performance evaluations demonstrate the effectiveness of our proposed algorithm with respect to the system-wide cost and the number of patients benefiting from MEC.

Due to line-of-sight communication links and distributed deployment, Unmanned Aerial Vehicles (UAVs) have attracted substantial interest in agile Mobile Edge Computing (MEC) service provision. In this paper, by clustering multiple users into independent communities based on their geographic locations, we design a 5G-enabled UAV-to-community offloading system. A system throughput maximization problem is formulated, subjected to the transmission rate, atomicity of tasks and speed of UAVs. By relaxing the transmission rate constraint, the mixed integer non-linear program is transformed into two subproblems. We first develop an average throughput maximization-based auction algorithm to determine the trajectory of UAVs, where a community-based latency approximation algorithm is developed to regulate the designed auction bidding. Then, a dynamic task admission algorithm is proposed to solve the task scheduling subproblem within one community. Performance analyses demonstrate that our designed auction bidding can guarantee user truthfulness, and can be fulfilled in polynomial time. Extensive simulations based on real-world data in health monitoring and online YouTube video services show that our proposed algorithm is able to maximize the system throughput while guaranteeing the fraction of served users.

The explosive growth of mobile devices promotes the prosperity of novel mobile applications, which can be realized by service offloading with the assistance of edge computing servers. However, due to limited computation and storage capabilities of a single server, long service latency hinders the continuous development of service offloading in mobile networks. By supporting multi-server cooperation, Pervasive Edge Computing (PEC) is promising to enable service migration in highly dynamic mobile networks. With the objective of maximizing the system utility, we formulate the optimization problem by jointly considering the constraints of server storage capability and service execution latency. To enable dynamic service placement, we first utilize Lyapunov optimization method to decompose the long-term optimization problem into a series of instant optimization problems. Then, a sample average approximation-based stochastic algorithm is proposed to approximate the future expected system utility. Afterwards, a distributed Markov approximation algorithm is utilized to determine the service placement configurations. Through theoretical analysis, the time complexity of our proposed algorithm is linear to the number of users, and the backlog queue of PEC servers is stable. Performance evaluations are conducted based on both synthetic and real trace-driven scenarios, with numerical results demonstrating the effectiveness of our proposed algorithm from various aspects.

The emerging 5G-enabled vehicular networks can satisfy various requirements of vehicles by traffic offloading. However, limited cellular spectrum and energy supplies restrict the development of 5G-enabled applications in vehicular networks. In this article, we construct an intelligent offloading framework for 5G-enabled vehicular networks, by jointly utilizing licensed cellular spectrum and unlicensed channels. A cost minimization problem is formulated by considering the latency constraint of users and is further decomposed into two subproblems due to its complexity. For the first subproblem, a two-sided matching algorithm is proposed to schedule the unlicensed spectrum. Then, a deep-reinforcement-learning-based method is investigated for the second one, where the system state is simplified to realize distributed traffic offloading. Real-world traces of taxies are leveraged to illustrate the effectiveness of our solution.

With the evolutionary development of latency sensitive applications, delay restriction is becoming an obstacle to run sophisticated applications on mobile devices. Partial computation offloading is promising to enable these applications to execute on mobile user equipments with low latency. However, most of the existing researches focus on either cloud computing or mobile edge computing (MEC) to offload tasks. In this paper, we comprehensively consider both of them and it is an early effort to study the cooperation of cloud computing and MEC in Internet of Things. We start from the single user computation offloading problem, where the MEC resources are not constrained. It can be solved by the branch and bound algorithm. Later on, the multiuser computation offloading problem is formulated as a mixed integer linear programming problem by considering resource competition among mobile users, which is NP-hard. Due to the computation complexity of the formulated problem, we design an iterative heuristic MEC resource allocation algorithm to make the offloading decision dynamically. Simulation results demonstrate that our algorithm outperforms the existing schemes in terms of execution latency and offloading efficiency.

The development of smart vehicles brings drivers and passengers a comfortable and safe environment. Various emerging applications are promising to enrich users’ traveling experiences and daily life. However, how to execute computing-intensive applications on resource-constrained vehicles still faces huge challenges. In this article, we construct an intelligent offloading system for vehicular edge computing by leveraging deep reinforcement learning. First, both the communication and computation states are modelled by finite Markov chains. Moreover, the task scheduling and resource allocation strategy is formulated as a joint optimization problem to maximize users’ Quality of Experience (QoE). Due to its complexity, the original problem is further divided into two sub-optimization problems. A two-sided matching scheme and a deep reinforcement learning approach are developed to schedule offloading requests and allocate network resources, respectively. Performance evaluations illustrate the effectiveness and superiority of our constructed system.