The Large Language Model (LLM) watermark is a newly emerging technique that shows promise in addressing concerns surrounding LLM copyright, monitoring AI-generated text, and preventing its misuse. The LLM watermark scheme commonly includes generating secret keys to partition the vocabulary into green and red lists, applying a perturbation to the logits of tokens in the green list to increase their sampling likelihood, thus facilitating watermark detection to identify AI-generated text if the proportion of green tokens exceeds a threshold. However, recent research indicates that watermarking methods using numerous keys are susceptible to removal attacks, such as token editing, synonym substitution, and paraphrasing, with robustness declining as the number of keys increases. Therefore, the state-of-the-art watermark schemes that employ fewer or single keys have been demonstrated to be more robust against text editing and paraphrasing. In this paper, we propose a novel green list stealing attack against the state-of-the-art LLM watermark scheme and systematically examine its vulnerability to this attack. We formalize the attack as a mixed integer programming problem with constraints. We evaluate our attack under a comprehensive threat model, including an extreme scenario where the attacker has no prior knowledge, lacks access to the watermark detector API, and possesses no information about the LLM’s parameter settings or watermark injection/detection scheme. Extensive experiments on LLMs, such as OPT and LLaMA, demonstrate that our attack can successfully steal the green list and remove the watermark across all settings.

Adversarial attacks are a serious threat to the reliable deployment of machine learning models in safety-critical applications. They can misguide current models to predict incorrectly by slightly modifying the inputs. Recently, substantial work has shown that adversarial examples tend to deviate from the underlying data manifold of normal examples, whereas pre-trained masked language models can fit the manifold of normal NLP data. To explore how to use the masked language model in adversarial detection, we propose a novel textual adversarial example detection method, namely Masked Language Model-based Detection (MLMD), which can produce clearly distinguishable signals between normal examples and adversarial examples by exploring the changes in manifolds induced by the masked language model. MLMD features a plug and play usage (i.e., no need to retrain the victim model) for adversarial defense and it is agnostic to classification tasks, victim model’s architectures, and to-be-defended attack methods. We evaluate MLMD on various benchmark textual datasets, widely studied machine learning models, and state-of-the-art (SOTA) adversarial attacks (in total 3*4*4 = 48 settings). Experimental results show that MLMD can achieve strong performance, with detection accuracy up to 0.984, 0.967, and 0.901 on AG-NEWS, IMDB, and SST-2 datasets, respectively. Additionally, MLMD is superior, or at least comparable to, the SOTA detection defenses in detection accuracy and F1 score. Among many defenses based on the off-manifold assumption of adversarial examples, this work offers a new angle for capturing the manifold change. The code for this work is openly accessible at https://github.com/mlmddetection/MLMDdetection.

In mobile crowdsourcing, the accuracy of the collected data is usually hard to ensure. Researchers have proposed techniques to identify truth from noisy data by inferring and utilizing the reliability of mobile users, and allocate tasks to users with higher reliability. However, they neglect the fact that a user may only have expertise on some problems (in some domains), but not others, and hence causing two problems: low estimation accuracy in truth analysis and ineffective task allocation. To address these problems, we propose Expertise-aware Truth Analysis and Task Allocation (ETA2), which can effectively infer user expertise, and then estimate truth and allocate tasks based on the inferred expertise. ETA2relies on a novel semantic analysis method to identify the expertise, and an expertise-aware truth analysis method to find the truth. For expertise-aware task allocation in ETA2, we formalize and solve two problems based on the optimization objectives: max-qualitytask allocation which maximizes the probability fortasks to be allocated to users with high expertise and min-costtask allocation which minimizes the cost of task allocation while ensuring high-quality data are collected. Experimental results based on two real-world datasets and one synthetic dataset demonstrate that ETA2significantly outperforms existing solutions.

Mobile cloud computing has been widely used to support the computation-intensive applications in mobile devices. Since cloud computing relies on the facilities like network infrastructures and cloud servers, the cloud service can be easily hampered by the limitation of these facilities in many cases. Therefore, researchers proposed the concept of device-to-device offloading to offload workload to nearby devices, without going through other facilities. Most existing works on device-to-device offloading aim to minimize the execution time of tasks or minimize the energy consumption. These works assume devices can be connected for a long time, but they neglect the fact that the connections among mobile devices are usually intermittent and even opportunistic, which may render the failure of task offloading. In this paper, we propose a new approach PeerCloud that aims to improve the success ratio of task offloading. Specifically, since mobile devices are carried by human beings, we study the social relationship between the device carriers and discover the hidden regularity in their contact patterns, in order to predict the likelihood of node departure. An optimization problem is then formalized to maximize the success ratio of task offloading within tasks' time constraints. Experimental results based on two real-world datasets demonstrate that the PeerCloud outperforms existing approaches by significantly improving the success ratio.

Mobile opportunistic networks have been exploited for data forwarding and data offloading in many network scenarios, like the mobile edge networks, due to its low cost and high robustness. Existing data forwarding strategies exploit all available network resources to forward data in a ‘best‐effort’ manner. However, they ignore data's heterogeneous delay constraints and may ineffectively assign network resources, resulting in ineffective data forwarding. In this study, the authors improve the existing strategies by proposing a backbone‐based on‐demand data forwarding strategy, which assign network resources to data items on‐demand, according to their delay requirements. Specifically, they first propose an algorithm to extract a backbone structure in the network, where nodes in the backbone structure are responsible for the data forwarding in the whole network. Then, on‐demand data forwarding is formalised as an optimisation problem, which selects the minimum number of paths from the backbone to ensure data are delivered on time with high confidence. To address this problem, a path elimination process and a path selection algorithm are proposed to select highly‐independent paths according to the delay requirements of data. Evaluation results show that the proposed on‐demand strategy can significantly improve the performance of data forwarding in mobile opportunistic networks.

Mobile crowdsourcing has received considerable attention as it enables people to collect and share large volume of data through their mobile devices. Since the accuracy of the collected data is usually hard to ensure, researchers have proposed techniques to identify truth from noisy data by inferring and utilizing the reliability of users, and allocate tasks to users with higher reliability. However, they neglect the fact that a user may only have expertise on some problems (in some domains), but not others. Neglecting this expertise diversity may cause two problems: low estimation accuracy in truth analysis and ineffective task allocation. To address these problems, we propose an Expertise-aware Truth Analysis and Task Allocation (ETA2) approach, which can effectively infer user expertise and then allocate tasks and estimate truth based on the inferred expertise. ETA2relies on a novel semantic analysis method to identify the expertise domains of the tasks and user expertise, an expertise-aware truth analysis solution to estimate truth and learn user expertise, and an expertise-aware task allocation method to maximize the probability that tasks are allocated to users with the right expertise while ensuring the work load does not exceed the processing capability at each user. Experimental results based on two real-world datasets demonstrate that ETA2significantly outperforms existing solutions.

Community detection has received considerable attention because of its applications to many practical problems in mobile networks. However, when considering temporal information associated with a community (i.e., transient community), most existing community detection methods fail due to their aggregation of contact information into a single weighted or unweighted network. In this paper, we propose a contact-burst-based clustering method to detect transient communities by exploiting pairwise contact processes. In this method, we formulate each pairwise contact process as a regular appearance of contact bursts, during which most contacts between the pair of nodes happen. Based on this formulation, we detect transient communities by clustering the pairs of nodes with similar contact bursts. Since it is difficult to collect global contact information at individual nodes, we further propose a distributed method to detect transient communities. In addition to transient community detection, we also propose a new data forwarding strategy for delay tolerant networks, in which transient communities serve as the data forwarding unit. Evaluation results show that our strategy can achieve a much higher data delivery ratio than traditional community-based strategies with comparable network overhead.

Photo crowdsourcing with smartphone has attracted considerable attention recently due to the prevalence of smartphones and the rich information provided by photos. In scenarios such as disaster recovery or battlefield, where the cellular network is partly damaged or severely overloaded, Disruption Tolerant Networks (DTNs) become the best way to deliver the crowdsourced photos. Since the bandwidth and storage resources in DTN are very limited and not enough to deliver all the crowdsourced photos, it is important to prioritize more valuable photos to use the limited resources. In this paper, we design a resource-aware photo crowdsourcing framework in DTN, which uses photo metadata including the smartphone's location, orientation, and other built-in camera's parameters, to estimate the value of photos. We propose a photo selection algorithm to maximize the value of photos delivered to the command center considering bandwidth and storage constraints. Both prototype implementation and trace-driven simulations demonstrate the effectiveness of our design.

Although crowdsourcing can provide a large amount of information through mobile devices and mobile users, the information provided by them may be inaccurate. Various truth analysis techniques have been proposed to identify truth from the noisy data either in a heuristic manner or using statistical models. However, if the available data are limited or have large conflicts, it is difficult to identify the truth or ensure the data credibility (quality). In this paper, we address this problem by utilizing the communication networks to adaptively collect data from mobile users, especially when the existing data are not enough to ensure data credibility. Considering the requirement on data credibility and the constraint of network resources, we quantify the tradeoff between the enhanced data credibility and the increased network overhead, and propose resource-aware approaches for truth analysis. Specifically, we formalize two problems in resource-constrained mobile opportunistic networks: max-credibility which aims to maximize data credibility with some network overhead, and min-overhead which aims to achieve a specified data credibility while minimizing the network overhead. Simulation and experimental results demonstrate the effectiveness of the proposed solutions in terms of data credibility and network overhead.

Human mobility prediction has received considerable attention because it helps addressing many practical problems in mobile networks. Most existing techniques focus on regular mobility prediction by studying the periodic mobility pattern of users. However, they fail to detect users' irregular mobility patterns, like attending a sporadic event. We address this problem by proposing techniques to predict event attendance based on the following basic idea: if a user is interested in events related to a topic, he may also attend future events related to this topic. In our solution, to learn how users are likely to attend the future events, three sets of features are identified by analyzing users' past activities, including semantic, temporal, and spatial features. Then, the supervised learning models are trained to predict event attendance based on the extracted features. To evaluate the performance of the proposed techniques, we collect a dataset based on Meet up that contains semantic descriptions of all events organized over a period of two years. Evaluation results show that the supervised classifiers built by all features outperform those built by individual features, and semantic features are more effective than temporal features and spatial features for predicting event attendance.

Vaccination is one of the most effective ways to protect people from being infected by infectious disease. However, it is often impractical to vaccinate all people in a community due to various resource constraints. Therefore, targeted vaccination, which vaccinates a small group of people, is an alternative approach to contain infectious disease spread. To achieve better performance in targeted vaccination, we collect student contact traces in a high school based on wireless sensors carried by students. With our wireless sensor system, we can record student contacts within the disease propagation distance, and then construct a disease propagation graph to model the infectious disease propagation. Based on this graph, we propose a metric called connectivity centrality to measure a node's importance during disease propagation and design centrality based algorithms for targeted vaccination. The proposed algorithms are evaluated and compared with other schemes based on our collected traces. Trace driven simulation results show that our algorithms can help to effectively contain infectious disease.

Mobile Social Network (MSN) with diverse connectivity characteristics is a combination of opportunistic network and mobile ad hoc network. Since the major difficulty of data forwarding is the opportunistic part, techniques designed for opportunistic networks are commonly used to forward data in MSNs. However, this may not be the best solution since they do not consider the ubiquitous existences of Transient Connected Components (TCCs), where nodes inside a TCC can reach each other by multi-hop wireless communications. In this paper, we first identify the existence of TCCs and analyze their properties based on five real traces. Then, we propose TCC-aware data forwarding strategies which exploit the special characteristics of TCCs to increase the contact opportunities and then improve the performance of data forwarding. Trace-driven simulations show that our TCC-aware data forwarding strategies outperform existing data forwarding strategies in terms of data delivery ratio and network overhead.

In mobile opportunistic networks, most existing research focuses on how to choose appropriate relays to carry and forward data. Although relay selection is an important issue, other issues such as finding content from people with the right expertise are also very important since the ultimate goal of using mobile opportunistic network is to provide the right content to mobile users (nodes). In this paper, we study expertise-based data access in content-centric mobile opportunistic networks, where the objective is to minimize the average query delay given a sequence of queries considering node expertise, node queuing delay and communication delay. To solve this problem, we propose various query forwarding approaches under deterministic and probabilistic expertise models. Specifically, we propose centralized approaches to assign queries based on a modified Dijkstra's shortest path algorithm and distributed approaches in which query forwarding is based on a utility metric. Extensive simulations on both synthetic and realistic traces demonstrate that our solutions outperform existing approaches.

Community has received considerable attention because of its application to many practical problems in mobile networks. However, when considering temporal information associated with community (i.e., transient community), most existing community detection methods fail due to their aggregation of the contact information into a single weighted or unweighted network. In this paper, we propose a contact-burst-based clustering method to detect transient communities by exploiting the pairwise contact processes. In this method, we formulate each pairwise contact process as regular appearance of contact bursts, during which most contacts between the pair of nodes happen. Based on such formulation, we detect transient communities by clustering the pairs of nodes with similar contact bursts together. We also propose a new data forwarding strategy for delay tolerant networks in which transient communities serve as the data forwarding unit. Evaluation results show that our strategy can achieve much higher data delivery ratio than traditional community-based strategies with comparable network overhead.