All publications

Activation space and parameter space provide complementary views of model computation. Activations represent information, while weights read, transform, and write that information. Yet existing interpretability methods largely study the two spaces separately, leaving the connection between represented information and parameter-level computation underexplored. We introduce Activation-Supported Parameter Decomposition (ASPD), which jointly decomposes activation and parameter spaces and grounds each learned weight component in the activation features it reads or writes. This grounding constrains otherwise non-unique parameter decompositions using the model's internal activations, while an internal reconstruction objective provides a local learning signal at the weight matrix being analyzed. Together, these properties enable scalable, interpretable, and causally editable parameter decomposition in pretrained large language models, demonstrated on Qwen-3-8B. The learned read--write components can also be composed into parameter-level mechanism circuits. We use ASPD to recover mechanisms underlying the classic IOI circuit and trace semantic transformations through model weights.

Existing interpretable graph additive models still face limitations in either computational scalability or modeling flexibility. In terms of structural modeling, previous approaches either face quadratic scaling costs or sacrifice explicit source-to-target contribution decomposition. In terms of feature components, they rely either on per-feature neural networks or on single shared bases with limited feature specialization. We address both problems by introducing HARMONIA: Interpretable Graph Learning through Mixtures of Neural Bases, an interpretable-by-design framework. For feature modeling, HARMONIA introduces a Mixture of Neural Bases (MoNB), which routes features to specialized basis experts, enabling parameter sharing without sacrificing feature-specific specialization. For structural modeling, HARMONIA uses Relative Random Walk Probabilities (RRWP) to capture multi-hop and multi-path relationships, and proposes Sparse RRWP Aggregation (SRA) to compute these interactions through sparse graph propagation without quadratic pairwise complexity. HARMONIA retains a simple additive form in which predictions decompose into feature responses modulated by structural influence. Empirically, HARMONIA achieves stronger explanation recovery than existing interpretable graph baselines while maintaining competitive predictive performance and scaling to graphs with millions of nodes. These results show that interpretable graph learning can remain both faithful and scalable without sacrificing predictive effectiveness.

Generating adversarial driving scenarios is critical for evaluating and improving autonomous vehicle decision-making systems in simulation. Recent approaches rely primarily on the prior knowledge of Large Language Models and Vision-Language Models to generate driving scenarios procedurally. We argue that adversarial scenes should be generated based on the failure diagnosis (e.g., indecisiveness, multi-frame inconsistency) of the driving policy to specifically address the policy's weaknesses instead of relying on prior assumptions. In this paper, we propose SPHINX, a closed-loop framework for adversarial scenario synthesis guided by a simple principle: first explain, then explore. Beyond blindly exploring the scenario space, SPHINX leverages explainable artificial intelligence methods to analyze the policy, identifying key visual concepts and their influence on policy outputs, and the uncertainty of the decisions. Given the interpretable evidence extracted from the policy's own decision process, we use a vision language model to rationalize and criticize failure modes of the current policy. These critics are then used to generate targeted adversarial scenarios for policy retraining and improvement. We demonstrate that SPHINX can highlight an interpretable account of policy failures while other adversarial scene generation cannot. Across the evaluated benchmarks and test suites, SPHINX can be applied to diverse state-of-the-art autonomous vehicle architectures and yields consistent robustness improvements over existing scenario-generation methods.

Learning hierarchical features in Sparse Autoencoders (SAEs) is essential for capturing the structured nature of real-world data and mitigating issues like feature absorption or splitting. Existing works attempt to identify hierarchical relationships within independent feature sets by relying on activation coverage, the assumption that child feature should only activate when its parent feature activates. However, we demonstrate that this condition alone is insufficient; that is, it often produces false positives where parent and child concepts are semantically unrelated. To address this, we introduce a novel reconstruction condition that enforces a deeper functional link between hierarchical levels. By combining both activation and reconstruction constraints, we propose the Tree SAE, a model designed to learn hierarchical structures directly from within the feature set. Our results demonstrate that Tree SAEs significantly surpass the existing SAEs at learning hierarchical pairs while maintaining competitive performance to the state-of-the-art on several key benchmarks. Finally, we demonstrate the practical utility of our Tree SAE in mapping the geometry of child feature subspaces and uncovering the complex hierarchical concept structures encoded within large language models.

Tabular data remains prevalent in high-stakes domains such as healthcare and finance, where predictive models are expected to provide both high accuracy and faithful, human-understandable reasoning. While symbolic models offer verifiable logic, they lack semantic expressiveness. Meanwhile, general-purpose LLMs often require specialized fine-tuning to master domain-specific tabular reasoning. To address the dual challenges of scalable data curation and reasoning consistency, we propose ReSS, a systematic framework that bridges symbolic and neural reasoning models. ReSS leverages a decision-tree model to extract instance-level decision paths as symbolic scaffolds. These scaffolds, alongside input features and labels, guide an LLM to generate grounded natural-language reasoning that strictly adheres to the underlying decision logic. The resulting high-quality dataset is used to fine-tune a pretrained LLM into a specialized tabular reasoning model, further enhanced by a scaffold-invariant data augmentation strategy to improve generalization and explainability. To rigorously assess faithfulness, we introduce quantitative metrics including hallucination rate, explanation necessity, and explanation sufficiency. Experimental results on medical and financial benchmarks demonstrate that ReSS-trained models improve traditional decision trees and standard fine-tuning approaches up to $10\%$ while producing faithful and consistent reasoning

Providing textual concept-based explanations for neurons in deep neural networks (DNNs) is of importance in understanding how a DNN model works. Prior works have associated concepts with neurons based on examples of concepts or a pre-defined set of concepts, thus limiting possible explanations to what the user expects, especially in discovering new concepts. Furthermore, defining the set of concepts requires manual work from the user, either by directly specifying them or collecting examples. To overcome these, we propose to leverage multimodal large language models for automatic and open-ended concept discovery. We show that, without a restricted set of pre-defined concepts, our method gives rise to novel interpretable concepts that are more faithful to the model's behavior. To quantify this, we validate each concept by generating examples and counterexamples and evaluating the neuron's response on this new set of images. Collectively, our method can discover concepts and simultaneously validate them, providing a credible automated tool to explain deep neural networks.

Temporal Graph Neural Network (TGNN) has been receiving a lot of attention recently due to its capability in modeling time-evolving graph-related tasks. Similar to Graph Neural Networks, it is also non-trivial to interpret predictions made by a TGNN due to its black-box nature. A major approach tackling this problems in GNNs is by analyzing the model' responses on some perturbations of the model's inputs, called perturbation-based explanation methods. While these methods are convenient and flexible since they do not need internal access to the model, does this lack of internal access prevent them from revealing some important information of the predictions? Motivated by that question, this work studies the limit of some classes of perturbation-based explanation methods. Particularly, by constructing some specific instances of TGNNs, we show (i) node-perturbation cannot reliably identify the paths carrying out the prediction, (ii) edge-perturbation is not reliable in determining all nodes contributing to the prediction and (iii) perturbing both nodes and edges does not reliably help us identify the graph's components carrying out the temporal aggregation in TGNNs.

Temporal graph neural networks (TGNNs) have been widely used for modeling time-evolving graph-related tasks due to their ability to capture both graph topology dependency and non-linear temporal dynamic. The explanation of TGNNs is of vital importance for a transparent and trustworthy model. However, the complex topology structure and temporal dependency make explaining TGNN models very challenging. In this paper, we propose a novel explainer framework for TGNN models. Given a time series on a graph to be explained, the framework can identify dominant explanations in the form of a probabilistic graphical model in a time period. Case studies on the transportation domain demonstrate that the proposed approach can discover dynamic dependency structures in a road network for a time period.

In many modern image-classification applications, understanding the cause of model's prediction can be as critical as the prediction's accuracy itself. Various feature-based local explanations generation methods have been designed to give us more insights on the decision of complex classifiers. Nevertheless, there is no consensus on evaluating the quality of different explanations. In response to this lack of comprehensive evaluation, we introduce the c-Eval metric and its corresponding framework to quantify the feature-based local explanation's quality. Given a classifier's prediction and the corresponding explanation on that prediction, c-Eval is the minimum-distortion perturbation that successfully alters the prediction while keeping the explanation's features unchanged. We then demonstrate how c-Eval can be computed using some modifications on existing adversarial generation libraries. To show that c-Eval captures the importance of input's features, we establish the connection between c-Eval and the features returned by explainers in affine and nearly-affine classifiers. We then introduce the c-Eval plot, which not only displays a strong connection between c-Eval and explainers' quality, but also helps automatically determine explainer's parameters. Since the generation of c-Eval relies on adversarial generation, we provide a demo of c-Eval on adversarial-robust models and show that the metric is applicable in those models. Finally, extensive experiments of explainers on different datasets are conducted to support the adoption of c-Eval in evaluating explainers' performance.

Knowledge Distillation (KD) has been considered as a key solution in model compression and acceleration in recent years. In KD, a small student model is generally trained from a large teacher model by minimizing the divergence between the probabilistic outputs of the two. However, as demonstrated in our experiments, existing KD methods might not transfer critical explainable knowledge of the teacher to the student, i.e. the explanations of predictions made by the two models are not consistent. In this paper, we propose a novel explainable knowledge distillation model, called XDistillation, through which both the performance the explanations' information are transferred from the teacher model to the student model. The XDistillation model leverages the idea of convolutional autoencoders to approximate the teacher explanations. Our experiments shows that models trained by XDistillation outperform those trained by conventional KD methods not only in term of predictive accuracy but also faithfulness to the teacher models.

In many modern image-classification applications, understanding the cause of model's prediction can be as critical as the prediction's accuracy itself. Various feature-based local explanations generation methods have been designed to give us more insights on the decision of complex classifiers. Nevertheless, there is no consensus on evaluating the quality of different explanations. In response to this lack of comprehensive evaluation, we introduce the c-Eval metric and its corresponding framework to quantify the feature-based local explanation's quality. Given a classifier's prediction and the corresponding explanation on that prediction, c-Eval is the minimum-distortion perturbation that successfully alters the prediction while keeping the explanation's features unchanged. We then demonstrate how c-Eval can be computed using some modifications on existing adversarial generation libraries. To show that c-Eval captures the importance of input's features, we establish the connection between c-Eval and the features returned by explainers in affine and nearly-affine classifiers. We then introduce the c-Eval plot, which not only displays a strong connection between c-Eval and explainers' quality, but also helps automatically determine explainer's parameters. Since the generation of c-Eval relies on adversarial generation, we provide a demo of c-Eval on adversarial-robust models and show that the metric is applicable in those models. Finally, extensive experiments of explainers on different datasets are conducted to support the adoption of c-Eval in evaluating explainers' performance.

Prompt tuning provides a parameter-efficient way to adapt foundation models (FMs) by freezing the pretrained backbone and updating only a small set of learnable prompts. This property makes prompt tuning especially suitable for decentralized federated learning (DFL), where exchanging full-model updates can be prohibitively expensive. However, prompt tuning in DFL introduces new challenges. Prompt sets learned from heterogeneous local data may not be index-wise aligned, making standard decentralized averaging unsuitable. In addition, the algorithm should be theoretically guaranteed to achieve consensus and make progress toward the shared objective. In this work, we provide the first study of prompt tuning in DFL. We formulate decentralized prompt tuning as a Wasserstein-based optimization problem over prompt measures, which captures the set-valued structure of prompts. We then propose D-FROST, an optimal-transport-based (OT-based) decentralized prompt-tuning algorithm that merges neighborhood prompts into compact representative prompt sets through transportation-based matching. We further analyze D-FROST by bounding the Wasserstein consensus error across clients, and establishing convergence of the network-level prompt barycenter to a neighborhood of stationarity. Experiments under heterogeneous client data demonstrate the effectiveness of D-FROST for decentralized prompt tuning.

Two-way partial AUC (TPAUC) is a critical performance metric for binary classification with imbalanced data, as it focuses on specific ranges of the true positive rate (TPR) and false positive rate (FPR). However, stochastic algorithms for TPAUC optimization remain under-explored, with existing methods either limited to approximated TPAUC loss functions or burdened by sub-optimal complexities. To overcome these limitations, we introduce two innovative stochastic primal-dual double block-coordinate algorithms for TPAUC maximization. These algorithms utilize stochastic block-coordinate updates for both the primal and dual variables, catering to both convex and non-convex settings. We provide theoretical convergence rate analyses, demonstrating significant improvements over prior approaches. Our experimental results, based on multiple benchmark datasets, validate the superior performance of our algorithms, showcasing faster convergence and better generalization. This work advances the state of the art in TPAUC optimization and offers practical tools for real-world machine learning applications.

Stochastic bilevel optimization finds widespread applications in machine learning, including meta-learning, hyperparameter optimization, and neural architecture search. To extend stochastic bilevel optimization to distributed data, several decentralized stochastic bilevel optimization algorithms have been developed. However, existing methods often suffer from slow convergence rates and high communication costs in heterogeneous settings, limiting their applicability to real-world tasks. To address these issues, we propose two novel decentralized stochastic bilevel gradient descent algorithms based on \textit{simultaneous} and \textit{alternating} update strategies. Our algorithms can achieve faster convergence rates and lower communication costs than existing methods. Importantly, our convergence analyses do not rely on strong assumptions regarding heterogeneity. More importantly, our theoretical analyses clearly disclose how the computation and communication regarding the Hessian-inverse-vector product under the heterogeneous setting affects the convergence rate. To the best of our knowledge, this is the first time such favorable theoretical results have been achieved with mild assumptions in the heterogeneous setting. Furthermore, we demonstrate how to establish the convergence rate for the alternating update strategy when combined with the variance-reduced gradient. Finally, experimental results confirm the efficacy of our algorithms.

Federated learning (FL) has emerged as a promising approach to collaboratively train machine learning models across multiple edge devices while preserving privacy. The success of FL hinges on the efficiency of participating models and their ability to handle the unique challenges of distributed learning. While several variants of Vision Transformer (ViT) have shown great potential as alternatives to modern convolutional neural networks (CNNs) for centralized training, the unprecedented size and higher computational demands hinder their deployment on resource-constrained edge devices, challenging their widespread application in FL. Since client devices in FL typically have limited computing resources and communication bandwidth, models intended for such devices must strike a balance between model size, computational efficiency, and the ability to adapt to the diverse and non-IID data distributions encountered in FL. To address these challenges, we propose OnDev-LCT: Lightweight Convolutional Transformers for On-Device vision tasks with limited training data and resources. Our models incorporate image-specific inductive biases through the LCT tokenizer by leveraging efficient depthwise separable convolutions in residual linear bottleneck blocks to extract local features, while the multi-head self-attention (MHSA) mechanism in the LCT encoder implicitly facilitates capturing global representations of images. Extensive experiments on benchmark image datasets indicate that our models outperform existing lightweight vision models while having fewer parameters and lower computational demands, making them suitable for FL scenarios with data heterogeneity and communication bottlenecks.

Federated learning is a new learning paradigm for extracting knowledge from distributed data. Due to its favorable properties in preserving privacy and saving communication costs, it has been extensively studied and widely applied to numerous data analysis applications. However, most existing federated learning approaches concentrate on the centralized setting, which is vulnerable to a single-point failure. An alternative strategy for addressing this issue is the decentralized communication topology. In this article, we systematically investigate the challenges and opportunities when renovating decentralized optimization for federated learning. In particular, we discussed them from the model, data, and communication sides, respectively, which can deepen our understanding about decentralized federated learning.

With the rapid growth of threats, sophistication and diversity in the manner of intrusion, traditional belt barrier systems are now faced with a major challenge of providing high and concrete coverage quality to expand the guarding service market. Recent efforts aim at constructing a belt barrier by deploying bistatic radar(s) on a specific line regardless of the limitation on deployment locations, to keep the width of the barrier from going below a specific threshold and the total bistatic radar placement cost is minimized, referred to as the Minimum Cost Linear Placement (MCLP) problem. The existing solutions are heuristic, and their validity is tightly bound by the barrier width parameter that these solutions only work for a fixed barrier width value. In this work, we propose an optimal solution, referred to as the Opt_MCLP, for the "open MCLP problem" that works for full range of the barrier width. Through rigorous theoretical analysis and experimentation, we demonstrate that the proposed algorithms perform well in terms of placement cost reduction and barrier coverage guarantee.

5G radio access network (RAN) with network slicing methodology plays a key role in the development of the next-generation network system. RAN slicing focuses on splitting the substrate's resources into a set of self-contained programmable RAN slices. Leveraged by network function virtualization (NFV), a RAN slice is constituted by various virtual network functions (VNFs) and virtual links that are embedded as instances on substrate nodes. In this work, we focus on the following fundamental tasks: i) establishing the theoretical foundation for constructing a VNF mapping plan for RAN slice recovery optimization and ii) developing algorithms needed to map/embed VNFs efficiently. In particular, we propose four efficient algorithms, including Resource-based Algorithm (RBA), Connectivity-based Algorithm (CBA), Group-based Algorithm (GBA), and Group-Connectivity-based Algorithm (GCBA) to solve the resource allocation and VNF mapping problem. Extensive experiments are also conducted to validate the robustness of RAN slicing via the proposed algorithms.

This paper studies a Group Influence with Minimum cost which aims to find a seed set with smallest cost that can influence all target groups, where each user is associated with a cost and a group is influenced if the total score of the influenced users belonging to the group is at least a certain threshold. As the group-influence function is neither submodular nor supermodular, theoretical bounds on the quality of solutions returned by the well-known greedy approach may not be guaranteed. To address this challenge, we propose a bi-criteria polynomial-time approximation algorithm with high certainty. At the heart of the algorithm is a novel group reachable reverse sample concept, which helps speed up the estimation of the group influence function. Finally, extensive experiments conducted on real social networks show that our proposed algorithm outperform the state-of-the-art algorithms in terms of the objective value and the running time.

Studying on networked systems, in which a communication between nodes is functional if their distance under a given metric is lower than a pre-defined threshold, has received significant attention recently. In this work, we propose a metric to measure network resilience on guaranteeing the pre-defined performance constraint. This metric is investigated under an optimization problem, namely \textbf{Length-bounded Paths Interdiction in Continuous Domain} (cLPI), which aims to identify a minimum set of nodes whose changes cause routing paths between nodes become undesirable for the network service. We show the problem is NP-hard and propose a framework by designing two oracles, \textit{Threshold Blocking} (TB) and \textit{Critical Path Listing} (CPL), which communicate back and forth to construct a feasible solution to cLPI with theoretical bicriteria approximation guarantees. Based on this framework, we propose two solutions for each oracle. Each combination of one solution to \tb and one solution to \cpl gives us a solution to cLPI. The bicriteria guarantee of our algorithms allows us to control the solutions's trade-off between the returned size and the performance accuracy. New insights into the advantages of each solution are further discussed via experimental analysis.

Graph mining plays a pivotal role across a number of disciplines, and a variety of algorithms have been developed to answer who/what type questions. For example, what items shall we recommend to a given user on an e-commerce platform? The answers to such questions are typically returned in the form of a ranked list, and graph-based ranking methods are widely used in industrial information retrieval settings. However, these ranking algorithms have a variety of sensitivities, and even small changes in rank can lead to vast reductions in product sales and page hits. As such, there is a need for tools and methods that can help model developers and analysts explore the sensitivities of graph ranking algorithms with respect to perturbations within the graph structure. In this paper, we present a visual analytics framework for explaining and exploring the sensitivity of any graph-based ranking algorithm by performing perturbation-based what-if analysis. We demonstrate our framework through three case studies inspecting the sensitivity of two classic graph-based ranking algorithms (PageRank and HITS) as applied to rankings in political news media and social networks.

Cost-aware Targeted Viral Marketing (CTVM), a generalization of Influence Maximization (IM), has received a lot of attentions recently due to its commercial values. Previous approximation algorithms for this problem required a large number of samples to ensure approximate guarantee. In this paper, we propose an efficient approximation algorithm which uses fewer samples but provides the same theoretical guarantees based on generating and using important samples in its operation. Experiments on real social networks show that our proposed method outperforms the state-of-the-art algorithm which provides the same approximation ratio in terms of the number of required samples and running time.

A major challenge in blockchain sharding protocols is that more than 95% transactions are cross-shard. Not only those cross-shard transactions degrade the system throughput but also double the confirmation time, and exhaust an already scarce network bandwidth. Are cross-shard transactions imminent for sharding schemes? In this paper, we propose a new sharding paradigm, called OptChain, in which cross-shard transactions are minimized, resulting in almost twice faster confirmation time and throughput. By treating transactions as a stream of nodes in an online graph, OptChain utilizes a lightweight and on-the-fly transaction placement method to group both related and soon-related transactions into the same shards. At the same time, OptChain maintains a temporal balance among shards to guarantee the high parallelism. Our comprehensive and large-scale simulation using Oversim P2P library confirms a significant boost in performance with up to 10 folds reduction in cross-shard transactions, more than twice reduction in confirmation time, and 50% increase in throughput. When combined with Omniledger sharding protocol, OptChain delivers a 6000 transactions per second throughput with 10.5s confirmation time.

This paper focuses on network resilience to perturbation of edge weight. Other than connectivity, many network applications nowadays rely upon some measure of network distance between a pair of connected nodes. In these systems, a metric related to network functionality is associated to each edge. A pair of nodes only being functional if the weighted, shortest-path distance between the pair is below a given threshold \texttt{T}. Consequently, a natural question is on which degree the change of edge weights can damage the network functionality? With this motivation, we study a new problem, \textit{Quality of Service Degradation}: given a set of pairs, find a minimum budget to increase the edge weights which ensures the distance between each pair exceeds $\mathtt{T}$. We introduce four algorithms with theoretical performance guarantees for this problem. Each of them has its own strength in trade-off between effectiveness and running time, which are illustrated both in theory and comprehensive experimental evaluation.

Given a weighted hypergraph $\mathcal{H}(V, \mathcal{E} \subseteq 2^V, w)$, the approximate $k$-cover problem seeks for a size-$k$ subset of $V$ that has the maximum weighted coverage by \emph{sampling only a few hyperedges} in $\mathcal{E}$. The problem has emerged from several network analysis applications including viral marketing, centrality maximization, and landmark selection. Despite many efforts, even the best approaches require $O(k n \log n)$ space complexities, thus, cannot scale to, nowadays, humongous networks without sacrificing formal guarantees. In this paper, we propose BCA, a family of algorithms for approximate $k$-cover that can find $(1-\frac{1}{e} -ε)$-approximation solutions within an \emph{$O(ε^{-2}n \log n)$ space}. That is a factor $k$ reduction on space comparing to the state-of-the-art approaches with the same guarantee. We further make BCA more efficient and robust on real-world instances by introducing a novel adaptive sampling scheme, termed DTA.

In this paper, a novel economic approach, based on the framework of contract theory, is proposed for providing incentives for LTE over unlicensed channels (LTE-U) in cellular-networks. In this model, a mobile network operator (MNO) designs and offers a set of contracts to the users to motivate them to accept being served over the unlicensed bands. A practical model in which the information about the quality-of-service (QoS) required by every user is not known to the MNO and other users, is considered. For this contractual model, the closed-form expression of the price charged by the MNO for every user is derived and the problem of spectrum allocation is formulated as a matching game with incomplete information. For the matching problem, a distributed algorithm is proposed to assign the users to the licensed and unlicensed spectra. Simulation results show that the proposed pricing mechanism can increase the fraction of users that achieve their QoS requirements by up to 45\% compared to classical algorithms that do not account for users requirements. Moreover, the performance of the proposed algorithm in the case of incomplete information is shown to approach the performance of the same mechanism with complete information.

Motivated by online social networks that are linked together through overlapping users, we study the influence maximization problem on a multiplex, with each layer endowed with its own model of influence diffusion. This problem is a novel version of the influence maximization problem that necessitates new analysis incorporating the type of propagation on each layer of the multiplex. We identify a new property, generalized deterministic submodular, which when satisfied by the propagation in each layer, ensures that the propagation on the multiplex overall is submodular -- for this case, we formulate ISF, the greedy algorithm with approximation ratio $(1 - 1/e)$. Since the size of a multiplex comprising multiple OSNs may encompass billions of users, we formulate an algorithm KSN that runs on each layer of the multiplex in parallel. KSN takes an $α$-approximation algorithm A for the influence maximization problem on a single-layer network as input, and has approximation ratio $\frac{(1-ε)α}{(o+1)k}$ for arbitrary $ε> 0$, $o$ is the number of overlapping users, and $k$ is the number of layers in the multiplex. Experiments on real and synthesized multiplexes validate the efficacy of the proposed algorithms for the problem of influence maximization in the heterogeneous multiplex. Implementations of ISF and KSN are available at http://www.alankuhnle.com/papers/mim/mim.html.

Device-to-device (D2D) communication is a promising approach to efficiently disseminate critical or viral information. Reaping the benefits of D2D-enabled networks is contingent upon choosing the optimal content dissemination policy subject to resource and user distribution constraints. In this paper, a novel D2D network analysis framework is proposed to study the impacts of frequently occurring subgraphs, known as motifs, on D2D network performance and to determine an effective content dissemination strategy. In the proposed framework, the distribution of devices in the D2D network is modeled as a Thomas cluster process (TCP), and two graph structures, the star and chain motifs, are studied in the communication graph. Based on the properties of the TCP, closed-form analytical expressions for the statistical significance, the outage probability, as well as the average throughput per device, are derived. Simulation results corroborate the analytical derivations and show the influence of different system topologies on the occurrence of motifs and the D2D system throughput. More importantly, the results highlight that, as the statistical significance of motifs increases, the system throughput will initially increase, and, then, decrease. Hence, network operators can obtain statistical significance regions for chain and star motifs that map to the optimal content dissemination performance. Furthermore, using the obtained regions and the analytical expressions for statistical significance, network operators can effectively identify which clusters of devices can be leveraged for D2D communications while determining the number of serving devices in each identified cluster.

While greedy algorithms have long been observed to perform well on a wide variety of problems, up to now approximation ratios have only been known for their application to problems having submodular objective functions $f$. Since many practical problems have non-submodular $f$, there is a critical need to devise new techniques to bound the performance of greedy algorithms in the case of non-submodularity. Our primary contribution is the introduction of a novel technique for estimating the approximation ratio of the greedy algorithm for maximization of monotone non-decreasing functions based on the curvature of $f$ without relying on the submodularity constraint. We show that this technique reduces to the classical $(1 - 1/e)$ ratio for submodular functions. Furthermore, we develop an extension of this ratio to the adaptive greedy algorithm, which allows applications to non-submodular stochastic maximization problems. This notably extends support to applications modeling incomplete data with uncertainty.

While greedy algorithms have long been observed to perform well on a wide variety of problems, up to now approximation ratios have only been known for their application to problems having submodular objective functions $f$. Since many practical problems have non-submodular $f$, there is a critical need to devise new techniques to bound the performance of greedy algorithms in the case of non-submodularity. Our primary contribution is the introduction of a novel technique for estimating the approximation ratio of the greedy algorithm for maximization of monotone non-decreasing functions based on the curvature of $f$ without relying on the submodularity constraint. We show that this technique reduces to the classical $(1 - 1/e)$ ratio for submodular functions. Furthermore, we develop an extension of this ratio to the adaptive greedy algorithm, which allows applications to non-submodular stochastic maximization problems. This notably extends support to applications modeling incomplete data with uncertainty.

Information can propagate among Online Social Network (OSN) users at a high speed, which makes the OSNs become important platforms for viral marketing. Although the viral marketing related problems in OSNs have been extensively studied in the past decade, the existing works all assume known propagation rates and are not able to solve the scenario when the rates may dynamically increase for popular topics. In this paper, we propose a novel model, Dynamic Influence Propagation (DIP), which allows propagation rates to change during the diffusion and can be used for describing information propagation in OSNs more realistically. Based on DIP, we define a new research problem: Threshold Activation Problem under DIP (TAP-DIP). TAP-DIP is more generalized than TAP and can be used for studying the DIP model. However, it adds another layer of complexity over the already \#P-hard TAP problem. Despite it hardness, we are able to approximate TAP-DIP with $O(\log|V|)$ ratio. Our solution consists of two major parts: 1) the Lipschitz optimization technique and 2) a novel solution to the general version of TAP, the Multi-TAP problem. We experimentally test our solution Using various real OSN datasets, and demonstrate that our solution not only generates high-quality yet much smaller seed sets when being aware of the rate increase, but also is scalable. In addition, considering DIP or not has a significant difference in seed set selection.

In this paper, we study how rumors in Online Social Networks (OSNs) may impact the performance of device-to-device (D2D) communication. As D2D is a new technology, people may choose not to use it when believed in rumors of its negative impacts. Thus, the cellular network with underlaying D2D is vulnerable to OSNs as rumors in OSNs may decrement the throughput of the cellular network in popular content delivery scenarios. To analyze the vulnerability, we introduce the problem of finding the most critical nodes in the OSN such that the throughput of a content delivery scenario is minimized when a rumor starts from those nodes. We then propose an efficient solution to the critical nodes detection problem. The severity of such vulnerability is supported by extensive experiments in various simulation settings, from which we observe up to $40\%$ reduction in network throughput.

Robustness in response to unexpected events is always desirable for real-world networks. To improve the robustness of any networked system, it is important to analyze vulnerability to external perturbation such as random failures or adversarial attacks occurring to elements of the network. In this paper, we study an emerging problem in assessing the robustness of complex networks: the vulnerability of the clustering of the network to the failure of network elements. Specifically, we identify vertices whose failures will critically damage the network by degrading its clustering, evaluated through the average clustering coefficient. This problem is important because any significant change made to the clustering, resulting from element-wise failures, could degrade network performance such as the ability for information to propagate in a social network. We formulate this vulnerability analysis as an optimization problem, prove its NP-completeness and non-monotonicity, and we offer two algorithms to identify the vertices most important to clustering. Finally, we conduct comprehensive experiments in synthesized social networks generated by various well-known models as well as traces of real social networks. The empirical results over other competitive strategies show the efficacy of our proposed algorithms.

Device-to-device (D2D) communications over licensed wireless spectrum has been recently proposed as a promising technology to meet the capacity crunch of next generation cellular networks. However, due to the high mobility of cellular devices, establishing and ensuring the success of D2D transmission becomes a major challenge. To this end, in this paper, a novel framework is proposed to enable devices to form multi-hop D2D connections in an effort to maintain sustainable communication in the presence of device mobility. To solve the problem posed by device mobility, in contrast to existing works, which mostly focus on physical domain information, a durable community based approach is introduced taking social encounters into context. It is shown that the proposed scheme can derive an optimal solution for time sensitive content transmission while also minimizing the cost that the base station pays in order to incentivize users to participate in D2D. Simulation results show that the proposed social community aware approach yields significant performance gain, in terms of the amount of traffic offloaded from the cellular network to the D2D tier, compared to the classical social-unaware methods.

Many networks including social networks, computer networks, and biological networks are found to divide naturally into communities of densely connected individuals. Finding community structure is one of fundamental problems in network science. Since Newman's suggestion of using \emph{modularity} as a measure to qualify the goodness of community structures, many efficient methods to maximize modularity have been proposed but without a guarantee of optimality. In this paper, we propose two polynomial-time algorithms to the modularity maximization problem with theoretical performance guarantees. The first algorithm comes with a \emph{priori guarantee} that the modularity of found community structure is within a constant factor of the optimal modularity when the network has the power-law degree distribution. Despite being mainly of theoretical interest, to our best knowledge, this is the first approximation algorithm for finding community structure in networks. In our second algorithm, we propose a \emph{sparse metric}, a substantially faster linear programming method for maximizing modularity and apply a rounding technique based on this sparse metric with a \emph{posteriori approximation guarantee}. Our experiments show that the rounding algorithm returns the optimal solutions in most cases and are very scalable, that is, it can run on a network of a few thousand nodes whereas the LP solution in the literature only ran on a network of at most 235 nodes.

Differentially private (DP) fine-tuning methods treat sparse Mixture-of-Experts (MoE) models as a single dense block, ignoring that shared layers see all data while experts only see routed records. We identify and formally characterize three resulting failure modes: global clipping suppresses expert gradients, batch-level normalization dilutes sparse expert updates, and fixed privacy noise degrades signal-to-noise ratio on low-load experts. We introduce RAPTOR - a Role-Aware Private Training framework, which alternates shared and expert optimization and targets each failure directly, using expert-specific clipping and noise together with a public expected-owner denominator and a count-independent update schedule that avoids conditioning on private, realized expert counts. We prove the resulting mechanism satisfies $(\varepsilon,δ)$-DP: because each record is assigned to exactly one owner expert, per-expert mechanisms within a layer compose in parallel, so updating all $E$ experts costs no more, in privacy terms, than updating one, with shared and expert streams composing sequentially across training. We further derive a bias-variance decomposition of the public-denominator estimator showing its bias grows predictably with routing imbalance, yielding a privacy-free rule for selecting which layer to protect from routing entropy measured on a small public corpus. Experiments on Switch Transformer and OLMoE fine-tuning across GLUE tasks, and on DeepSeek-VL2-Tiny, show consistent gains over standard DP baselines across several privacy levels ($\varepsilon$), with the largest margins typically at the tightest budgets. Code and models are publicly available: https://github.com/leduckhai/RAPTOR

Large Language Models (LLMs) have transformed natural language processing, demonstrating impressive capabilities across diverse tasks. However, deploying these models introduces critical risks related to intellectual property violations and potential misuse, particularly as adversaries can imitate these models to steal services or generate misleading outputs. We specifically focus on model stealing attacks, as they are highly relevant to proprietary LLMs and pose a serious threat to their security, revenue, and ethical deployment. While various watermarking techniques have emerged to mitigate these risks, it remains unclear how far the community and industry have progressed in developing and deploying watermarks in LLMs. To bridge this gap, we aim to develop a comprehensive systematization for watermarks in LLMs by 1) presenting a detailed taxonomy for watermarks in LLMs, 2) proposing a novel intellectual property classifier to explore the effectiveness and impacts of watermarks on LLMs under both attack and attack-free environments, 3) analyzing the limitations of existing watermarks in LLMs, and 4) discussing practical challenges and potential future directions for watermarks in LLMs. Through extensive experiments, we show that despite promising research outcomes and significant attention from leading companies and community to deploy watermarks, these techniques have yet to reach their full potential in real-world applications due to their unfavorable impacts on model utility of LLMs and downstream tasks. Our findings provide an insightful understanding of watermarks in LLMs, highlighting the need for practical watermarks solutions tailored to LLM deployment.

This paper introduces FairDP, a novel training mechanism designed to provide group fairness certification for the trained model's decisions, along with a differential privacy (DP) guarantee to protect training data. The key idea of FairDP is to train models for distinct individual groups independently, add noise to each group's gradient for data privacy protection, and progressively integrate knowledge from group models to formulate a comprehensive model that balances privacy, utility, and fairness in downstream tasks. By doing so, FairDP ensures equal contribution from each group while gaining control over the amount of DP-preserving noise added to each group's contribution. To provide fairness certification, FairDP leverages the DP-preserving noise to statistically quantify and bound fairness metrics. An extensive theoretical and empirical analysis using benchmark datasets validates the efficacy of FairDP and improved trade-offs between model utility, privacy, and fairness compared with existing methods. Our empirical results indicate that FairDP can improve fairness metrics by more than 65% on average while attaining marginal utility drop (less than 4% on average) under a rigorous DP-preservation across benchmark datasets compared with existing baselines.

Federated Learning (FL) has garnered significant attention for its potential to protect user privacy while enhancing model training efficiency. For that reason, FL has found its use in various domains, from healthcare to industrial engineering, especially where data cannot be easily exchanged due to sensitive information or privacy laws. However, recent research has demonstrated that FL protocols can be easily compromised by active reconstruction attacks executed by dishonest servers. These attacks involve the malicious modification of global model parameters, allowing the server to obtain a verbatim copy of users' private data by inverting their gradient updates. Tackling this class of attack remains a crucial challenge due to the strong threat model. In this paper, we propose a defense mechanism, namely OASIS, based on image augmentation that effectively counteracts active reconstruction attacks while preserving model performance. We first uncover the core principle of gradient inversion that enables these attacks and theoretically identify the main conditions by which the defense can be robust regardless of the attack strategies. We then construct our defense with image augmentation showing that it can undermine the attack principle. Comprehensive evaluations demonstrate the efficacy of the defense mechanism highlighting its feasibility as a solution.

Graph neural networks (GNNs) are susceptible to privacy inference attacks (PIAs), given their ability to learn joint representation from features and edges among nodes in graph data. To prevent privacy leakages in GNNs, we propose a novel heterogeneous randomized response (HeteroRR) mechanism to protect nodes' features and edges against PIAs under differential privacy (DP) guarantees without an undue cost of data and model utility in training GNNs. Our idea is to balance the importance and sensitivity of nodes' features and edges in redistributing the privacy budgets since some features and edges are more sensitive or important to the model utility than others. As a result, we derive significantly better randomization probabilities and tighter error bounds at both levels of nodes' features and edges departing from existing approaches, thus enabling us to maintain high data utility for training GNNs. An extensive theoretical and empirical analysis using benchmark datasets shows that HeteroRR significantly outperforms various baselines in terms of model utility under rigorous privacy protection for both nodes' features and edges. That enables us to defend PIAs in DP-preserving GNNs effectively.

In this paper, we show that the process of continually learning new tasks and memorizing previous tasks introduces unknown privacy risks and challenges to bound the privacy loss. Based upon this, we introduce a formal definition of Lifelong DP, in which the participation of any data tuples in the training set of any tasks is protected, under a consistently bounded DP protection, given a growing stream of tasks. A consistently bounded DP means having only one fixed value of the DP privacy budget, regardless of the number of tasks. To preserve Lifelong DP, we propose a scalable and heterogeneous algorithm, called L2DP-ML with a streaming batch training, to efficiently train and continue releasing new versions of an L2M model, given the heterogeneity in terms of data sizes and the training order of tasks, without affecting DP protection of the private training set. An end-to-end theoretical analysis and thorough evaluations show that our mechanism is significantly better than baseline approaches in preserving Lifelong DP. The implementation of L2DP-ML is available at: https://github.com/haiphanNJIT/PrivateDeepLearning.

Despite the great potential of Federated Learning (FL) in large-scale distributed learning, the current system is still subject to several privacy issues due to the fact that local models trained by clients are exposed to the central server. Consequently, secure aggregation protocols for FL have been developed to conceal the local models from the server. However, we show that, by manipulating the client selection process, the server can circumvent the secure aggregation to learn the local models of a victim client, indicating that secure aggregation alone is inadequate for privacy protection. To tackle this issue, we leverage blockchain technology to propose a verifiable client selection protocol. Owing to the immutability and transparency of blockchain, our proposed protocol enforces a random selection of clients, making the server unable to control the selection process at its discretion. We present security proofs showing that our protocol is secure against this attack. Additionally, we conduct several experiments on an Ethereum-like blockchain to demonstrate the feasibility and practicality of our solution.

In this paper, we focus on preserving differential privacy (DP) in continual learning (CL), in which we train ML models to learn a sequence of new tasks while memorizing previous tasks. We first introduce a notion of continual adjacent databases to bound the sensitivity of any data record participating in the training process of CL. Based upon that, we develop a new DP-preserving algorithm for CL with a data sampling strategy to quantify the privacy risk of training data in the well-known Averaged Gradient Episodic Memory (A-GEM) approach by applying a moments accountant. Our algorithm provides formal guarantees of privacy for data records across tasks in CL. Preliminary theoretical analysis and evaluations show that our mechanism tightens the privacy loss while maintaining a promising model utility.

The smart grid incentivizes distributed agents with local generation (e.g., smart homes, and microgrids) to establish multi-agent systems for enhanced reliability and energy consumption efficiency. Distributed energy trading has emerged as one of the most important multi-agent systems on the power grid by enabling agents to sell their excessive local energy to each other or back to the grid. However, it requests all the agents to disclose their sensitive data (e.g., each agent's fine-grained local generation and demand load). In this paper, to the best of our knowledge, we propose the first privacy preserving distributed energy trading framework, Private Energy Market (PEM), in which all the agents privately compute an optimal price for their trading (ensured by a Nash Equilibrium), and allocate pairwise energy trading amounts without disclosing sensitive data (via novel cryptographic protocols). Specifically, we model the trading problem as a non-cooperative Stackelberg game for all the agents (i.e., buyers and sellers) to determine the optimal price, and then derive the pairwise trading amounts. Our PEM framework can privately perform all the computations among all the agents without a trusted third party. We prove the privacy, individual rationality, and incentive compatibility for the PEM framework. Finally, we conduct experiments on real datasets to validate the effectiveness and efficiency of the PEM.

Although the iterative double auction has been widely used in many different applications, one of the major problems in its current implementations is that they rely on a trusted third party to handle the auction process. This imposes the risk of single point of failures, monopoly, and bribery. In this paper, we aim to tackle this problem by proposing a novel decentralized and trustless framework for iterative double auction based on blockchain. Our design adopts the smart contract and state channel technologies to enable a double auction process among parties that do not need to trust each other, while minimizing the blockchain transactions. In specific, we propose an extension to the original concept of state channels that can support multiparty computation. Then we provide a formal development of the proposed framework and prove the security of our design against adversaries. Finally, we develop a proof-of-concept implementation of our framework using Elixir and Solidity, on which we conduct various experiments to demonstrate its feasibility and practicality.

In this paper, we explore the partitioning attacks on the Bitcoin network, which is shown to exhibit spatial bias, and temporal and logical diversity. Through data-driven study we highlight: 1) the centralization of Bitcoin nodes across autonomous systems, indicating the possibility of BGP attacks, 2)the non-uniform consensus among nodes, that can be exploited to partition the network, and 3)the diversity in the Bitcoin software usage that can lead to privacy attacks. Atop the prior work, which focused on spatial partitioning, our work extends the analysis of the Bitcoin network to understand the temporal and logical effects on the robustness of the Bitcoin network.

Variational Quantum Algorithms (VQAs) are a leading paradigm for near-term quantum computing, combining parameterized quantum circuits with classical optimization across quantum chemistry, combinatorial optimization, and quantum machine learning. Since real-world VQA deployments routinely require circuits that exceed available hardware capacity, quantum circuit cutting has become an indispensable execution strategy, and pre-trained parameters are increasingly distributed through public repositories, introducing supply-chain security risks that have received little attention. Prior quantum backdoor attacks either introduce detectable circuit modifications or depend on device-specific noise, and none consider circuit cutting as an attack surface. We present CutBackdoor, the first parameter-supply-chain backdoor that uses cut circuit execution from CutQC as the deployment-time trigger against VQAs. Under noisy finite-shot circuit-cut execution, poisoned parameters preserve full-circuit validation performance while substantially increasing cut-path reconstruction error, without any circuit modification. The trigger activates when a resource-limited victim responds to a qubit-capacity mismatch by invoking the cutting workflow, requiring no attacker presence at deployment. We provide a theoretical analysis and empirically validate it across varying shot budgets. Evaluation across multiple VQA benchmarks on IBM quantum backends demonstrates cut-path energy amplification of $1.3\times$ to $2.9\times$ \revA{over clean baselines on the VQE and VQD benchmarks while maintaining small stealthiness error on the full-circuit path. The cut-path gap persists across the evaluated backends and cut placements under matched compilation; Zero-Noise Extrapolation provides only partial mitigation, and the diagonal-cost QAOA benchmark delineates the attack's structural boundary

Quantum Federated Learning (QFL) offers a promising framework to train quantum models across distributed clients while keeping data strictly local. Due to its simplicity and low communication overhead, Federated Averaging (FedAvg) is the standard aggregation choice in QFL literature. However, deploying QFL on practical hardware exposes a severe double-drift phenomenon: the global model is simultaneously derailed by client drift from non-IID data and hardware bias from noisy quantum gradient estimates. In this work, we first analyze the convergence of FedAvg under these realistic conditions, mathematically demonstrating that quantum hardware bias creates a persistent error floor that standard averaging cannot correct. To overcome this limitation, we propose Q-ANCHOR, a quantum-aware federated aggregation architecture that anchors server updates with zero-noise extrapolation while applying stateful client correction to suppress both client drift and hardware-induced bias. Our convergence theory proves that Q-ANCHOR successfully mitigates classical client drift while actively reducing the hardware-bias floor. Experimental results demonstrate that Q-ANCHOR achieves significantly more stable training than conventional FL baselines.

Quantum Machine Learning (QML) is becoming increasingly prevalent due to its potential to enhance classical machine learning (ML) tasks, such as classification. Although quantum noise is often viewed as a major challenge in quantum computing, it also offers a unique opportunity to enhance privacy. In particular, intrinsic quantum noise provides a natural stochastic resource that, when rigorously analyzed within the differential privacy (DP) framework and composed with classical mechanisms, can satisfy formal $(\varepsilon, δ)$-DP guarantees. This enables a reduction in the required classical perturbation without compromising the privacy budget, potentially improving model utility. However, the integration of classical and quantum noise for privacy preservation remains unexplored. In this work, we propose a hybrid noise-added mechanism, HYPER-Q, that combines classical and quantum noise to protect the privacy of QML models. We provide a comprehensive analysis of its privacy guarantees and establish theoretical bounds on its utility. Empirically, we demonstrate that HYPER-Q outperforms existing classical noise-based mechanisms in terms of adversarial robustness across multiple real-world datasets.

Network motif identification problem aims to find topological patterns in biological networks. Identifying non-overlapping motifs is a computationally challenging problem using classical computers. Quantum computers enable solving high complexity problems which do not scale using classical computers. In this paper, we develop the first quantum solution, called QOMIC (Quantum Optimization for Motif IdentifiCation), to the motif identification problem. QOMIC transforms the motif identification problem using a integer model, which serves as the foundation to develop our quantum solution. We develop and implement the quantum circuit to find motif locations in the given network using this model. Our experiments demonstrate that QOMIC outperforms the existing solutions developed for the classical computer, in term of motif counts. We also observe that QOMIC can efficiently find motifs in human regulatory networks associated with five neurodegenerative diseases: Alzheimers, Parkinsons, Huntingtons, Amyotrophic Lateral Sclerosis (ALS), and Motor Neurone Disease (MND).

Quantum annealing (QA) has emerged as a powerful technique to solve optimization problems by taking advantages of quantum physics. In QA process, a bottleneck that may prevent QA to scale up is minor embedding step in which we embed optimization problems represented by a graph, called logical graph, to Quantum Processing Unit (QPU) topology of quantum computers, represented by another graph, call hardware graph. Existing methods for minor embedding require a significant amount of running time in a large-scale graph embedding. To overcome this problem, in this paper, we introduce a novel notion of adaptive topology which is an expandable subgraph of the hardware graph. From that, we develop a minor embedding algorithm, namely Adaptive TOpology eMbedding (ATOM). ATOM iteratively selects a node from the logical graph, and embeds it to the adaptive topology of the hardware graph. Our experimental results show that ATOM is able to provide a feasible embedding in much smaller running time than that of the state-of-the-art without compromising the quality of resulting embedding.

Target Identification by Enzymes (TIE) problem aims to identify the set of enzymes in a given metabolic network, such that their inhibition eliminates a given set of target compounds associated with a disease while incurring minimum damage to the rest of the compounds. This is an NP-complete problem, and thus optimal solutions using classical computers fail to scale to large metabolic networks. In this paper, we consider the TIE problem for identifying drug targets in metabolic networks. We develop the first quantum optimization solution, called QuTIE (Quantum optimization for Target Identification by Enzymes), to this NP-complete problem. We do that by developing an equivalent formulation of the TIE problem in Quadratic Unconstrained Binary Optimization (QUBO) form, then mapping it to a logical graph, which is then embedded on a hardware graph on a quantum computer. Our experimental results on 27 metabolic networks from Escherichia coli, Homo sapiens, and Mus musculus show that QuTIE yields solutions which are optimal or almost optimal. Our experiments also demonstrate that QuTIE can successfully identify enzyme targets already verified in wet-lab experiments for 14 major disease classes.

Quantum annealing (QA) that encodes optimization problems into Hamiltonians remains the only near-term quantum computing paradigm that provides sufficient many qubits for real-world applications. To fit larger optimization instances on existing quantum annealers, reducing Hamiltonians into smaller equivalent Hamiltonians provides a promising approach. Unfortunately, existing reduction techniques are either computationally expensive or ineffective in practice. To this end, we introduce a novel notion of non-separable~group, defined as a subset of qubits in a Hamiltonian that obtains the same value in optimal solutions. We develop a theoretical framework on non-separability accordingly and propose FastHare, a highly efficient reduction method. FastHare, iteratively, detects and merges non-separable groups into single qubits. It does so within a provable worst-case time complexity of only $O(αn^2)$, for some user-defined parameter $α$. Our extensive benchmarks for the feasibility of the reduction are done on both synthetic Hamiltonians and 3000+ instances from the MQLIB library. The results show FastHare outperforms the roof duality, the implemented reduction method in D-Wave's SDK library, with 3.6x higher average reduction ratio. It demonstrates a high level of effectiveness with an average of 62% qubits saving and 0.3s processing time, advocating for Hamiltonian reduction as an inexpensive necessity for QA.

Understanding the COVID-19 vaccine hesitancy, such as who and why, is very crucial since a large-scale vaccine adoption remains as one of the most efficient methods of controlling the pandemic. Such an understanding also provides insights into designing successful vaccination campaigns for future pandemics. Unfortunately, there are many factors involving in deciding whether to take the vaccine, especially from the cultural point of view. To obtain these goals, we design a novel culture-aware machine learning (ML) model, based on our new data collection, for predicting vaccination willingness. We further analyze the most important features which contribute to the ML model's predictions using advanced AI explainers such as the Probabilistic Graphical Model (PGM) and Shapley Additive Explanations (SHAP). These analyses reveal the key factors that most likely impact the vaccine adoption decisions. Our findings show that Hispanic and African American are most likely impacted by cultural characteristics such as religions and ethnic affiliation, whereas the vaccine trust and approval influence the Asian communities the most. Our results also show that cultural characteristics, rumors, and political affiliation are associated with increased vaccine rejection.

Although the effects of the social norm on mitigating misinformation are identified, scant knowledge exists about the patterns of social norm emergence, such as the patterns and variations of social tipping in online communities with diverse characteristics. Accordingly, this study investigates the features of social tipping in online communities and examines the correlations between the tipping features and characteristics of online communities. Taking the side effects of COVID-19 vaccination as the case topic, we first track the patterns of tipping features in 100 online communities, which are detected using Louvain Algorithm from the aggregated communication network on Twitter between May 2020 and April 2021. Then, we use multi-variant linear regression to explore the correlations between tipping features and community characteristics. We find that social tipping in online communities can sustain for two to four months and lead to a 50% increase in populations who accept the normative belief in online communities. The regression indicates that the duration of social tipping is positively related to the community populations and original acceptance of social norms, while the correlation between the tipping duration and the degrees among community members is negative. Additionally, the network modularity and original acceptance of social norms have negative relationships with the extent of social tipping, while the degree and betweenness centrality can have significant positive relationships with the extent of tipping. Our findings shed light on more precise normative interventions on misinformation in digital environments as it offers preliminary evidence about the timing and mechanism of social norm emergence.

Social Networks (SNs) have been gradually applied by utility companies as an addition to smart grid and are proved to be helpful in smoothing load curves and reducing energy usage. However, SNs also bring in new threats to smart grid: misinformation in SNs may cause smart grid users to alter their demand, resulting in transmission line overloading and in turn leading to catastrophic impact to the grid. In this paper, we discuss the interdependency in the social network coupled smart grid and focus on its vulnerability. That is, how much can the smart grid be damaged when misinformation related to it diffuses in SNs? To analytically study the problem, we propose the Misinformation Attack Problem in Social-Smart Grid (MAPSS) that identifies the top critical nodes in the SN, such that the smart grid can be greatly damaged when misinformation propagates from those nodes. This problem is challenging as we have to incorporate the complexity of the two networks concurrently. Nevertheless, we propose a technique that can explicitly take into account information diffusion in SN, power flow balance and cascading failure in smart grid integratedly when evaluating node criticality, based on which we propose various strategies in selecting the most critical nodes. Also, we introduce controlled load shedding as a protection strategy to reduce the impact of cascading failure. The effectiveness of our algorithms are demonstrated by experiments on IEEE bus test cases as well as the Pegase data set.