Backdoor attacks pose a serious threat to the security of large language models (LLMs), causing them to exhibit anomalous behavior under specific trigger conditions. The design of backdoor triggers has evolved from fixed triggers to dynamic or implicit triggers. This increased flexibility in trigger design makes it challenging for defenders to accurately identify their specific forms. Most existing backdoor defense methods are limited to specific types of triggers or rely on an additional clean model for support. To address this issue, we propose a backdoor detection method based on attention similarity, enabling backdoor detection without prior knowledge of the trigger. Our study reveals that models subjected to backdoor attacks exhibit unusually high similarity among attention heads when exposed to triggers. Based on this observation, we propose an attention safety alignment approach combined with head-wise fine-tuning to rectify potentially contaminated attention heads, thereby effectively mitigating the impact of backdoor attacks. Extensive experimental results demonstrate that our method significantly reduces the success rate of backdoor attacks while preserving the model’s performance on downstream tasks.

Retrieval-Augmented Generation (RAG) techniques have emerged as a promising direction to merge the non-parametric knowledge into Large Language Models (LLMs), thereby alleviating factual errors, hallucinations and outdated knowledge. Existing RAG methods, which append multiple retrieved documents or passages to the input of LLMs, will inevitably increase the context length, resulting in not only significant computational overhead and inference latency, but also performance degradation. Although reranking or compression modules have been introduced to address these challenges, they overlook the contextual preferences of the generative LLMs itself and may inadvertently discard information that is crucial for generation accuracy. To this end, we introduce InnerRAG, which incentivizes RAG via Inner Adaptive Context Selection. InnerRAG is a novel paradigm that empowers LLMs to autonomously select relevant context during generation. Our proposed InnerRAG endows the model to accurately identify the documents that are most helpful for generation from long contexts. By endowing the model with this capability, InnerRAG facilitates more effective exploitation of external knowledge without being misled by disturbed information, leading to substantial improvements in generation quality while maintaining computational efficiency. Extensive experiments across multiple benchmarks and human evaluations demonstrate that our method consistently outperforms state-of-the-art RAG baselines. Moreover, our framework is orthogonal and complementary to in-context RAG approaches, offering further performance improvements when combined.

As image data outsourcing to mobile cloud grows, data privacy has become a major concern. Privacy-preserving image retrieval aims to search over encrypted data without requiring decryption. However, existing schemes face three critical challenges: struggle to balance accuracy, efficiency, and security; limited scalability for large-scale image retrieval in multi-user settings; and vulnerability to security threats arising from user permission changes or key compromises. In this paper, we propose ELSEIR, a novel framework for accurate, efficient, and privacy-preserving image retrieval. ELSEIR leverages a deep hashing model to extract image feature vectors and designs an irreversible random hash code generation module that combines secure permutation keys with two differential privacy methods for privacy protection. To enable accurate searches in multi-user settings, ELSEIR introduces a key conversion protocol that allows the cloud to unify ciphertexts encrypted under different user keys via corresponding switch keys. Furthermore, we extend the framework to ABE-ELSEIR, which supports immediate and efficient user revocation. We further provide formal security proofs demonstrating that the proposed frameworks are resilient against known-plaintext and key collusion attacks. Extensive experiments on real-world datasets show that our scheme achieves accuracy comparable to the unprotected baseline, while surpassing existing approaches in both retrieval accuracy and efficiency.

Backdoor attacks threaten the reliability of NLP models by embedding hidden behaviors during training, which are activated by specific inputs at inference time. Traditional backdoor triggers often rely on explicit content alterations-such as token insertion or stylistic modification-which may compromise semantic coherence and be easily detected.In this work, we propose a novel backdoor attack strategy that leverages the linguistic properties of code-mixing(a language form that combines elements from two or more languages) as implicit triggers. Drawing inspiration from natural code-mixing communication, we design three types of linguistically grounded triggers: inter-word mixing, intra-sentential mixing, and inter-sentential mixing. These forms reflect realistic language usage patterns in bilingual communities, enhancing the stealthiness of the attack. The experiment results show that existing NLP models perform poorly when faced with backdoor attacks based on code-mixing triggers. We are the first to focus on code-mixing as a trigger for text backdoor attacks. We hope this research raises awareness of the vulnerability of models during training when faced with code-mixing.

Retrieval-Augmented Image Captioning enhances the model's understanding of real-world images by retrieving external knowledge. Existing methods mainly use original captions or isolated entities related to the query image to help generate captions. However, these methods make the model either imitate the caption style or fail to capture the relationship between entities, resulting in a lack of diversity or inaccuracy in the generated captions. To address these issues, we propose SEAR, a novel framework that utilizes external Synthesized Entity-Aware knowledge Representations to improve captioning performance. Specifically, SEAR clusters images based on scene-level and entity-level features, and synthesizes each clustered images into representative images as retrieval indexes, and simultaneously utilizes a large model to extract and supplement structured knowledge graphs from the corresponding cluster captions. Furthermore, we design a knowledge-graph pruner to prune the knowledge graph by retaining the most relevant subgraphs to the query image. By undertaking these steps in an integrated manner, SEAR enables the model to acquire non-redundant and structured information for generating captions and avoid data-related privacy issues. Extensive experiments on MSCOCO, Flickr30k, and NoCaps demonstrate the effectiveness of our method both in-domain and out-of-domain, outperforming existing lightweight RAIC methods and remaining competitive with heavyweight models.

Structure-enhanced models get a leading performance on the link prediction task as they utilize selected structure features and Graph Neural Network (GNN) based node embeddings simultaneously. However, we observe that when graphs get sparser, these methods perform worse than classical GNN-based methods, which has a severe impact on their practical use. We prove that when the graph gets sparser, the distance between structure features gets smaller. We induce this is the underlying reason for hindering the model from giving a reasonable prediction and leading to performance degeneration. To overcome this problem, we first claim that models need to learn the importance of node embeddings based on the distance between structure features. However, there is a lack of research to efficiently estimate the distance information, and existing models fail to assign the importance properly. Then we design a method called DIP, which satisfies the relation requirement with a weighted term of node embeddings, and use node degree to estimate the distance information to get the weight. Experimental results show that DIP can significantly improve the accuracy of structure-enhanced link prediction models and solve the performance degeneration problem effectively. The code of DIP is publicly available at https://github.com/lzwqbh/DIP.

The remarkable image understanding and generation capabilities of diffusion models have made image editing a highly promising area of research. As a significant subtask within the field, object removal aims to remove objects from a specified region and fill the missing pixels with visually coherent and semantically sound content. Despite the great progress made in deep generative models, research in this area still faces several challenges: i. high expense of model training induced by the data scarcity and the difficulity in large-scale real data collection ii. current training-free methods are unable to drastically change the behavior of the attention layer that has been set during the pre-training phase for text-guided object inpainting. In this paper, we introduce HybridRemover, a two-stage diffusion scheme. We decouple the task into two subtasks: one to remove the specified objects from the target region and one to perform image restoration of the target region. By fine-tuning the SD-Inpainting model with a very small amount of data, we transform it into a model that focuses only on the complete removal of the object without considering the surrounding effects, and the overall repair of the image is taken care of by the SD-Inpainting model cascaded behind it. As a result of our efforts, our method achieved state-of-the-art performance in object removal tasks. Even when a strong perspective distortion gets involved, our method delivers exceptional results.

Efficiently managing data transmission across unreliable communication channels for diverse sources has attracted much attention due to its great practicability in the Internet of Things systems. Despite its significance in real-time wireless systems, ensuring stringent freshness requirements for each information source remains an understudied challenge, especially in scenarios demanding heterogeneous quality-of-service guarantees. In this paper, we study a scheduling problem of maximizing the accumulated information subject to different Age of Information (AoI) requirements for heterogeneous information sources. To that end, we first present a global optimal transmission scheduling policy when the source statistics are known a priori. Then, as this information is unknown, we formulate the problem as an AoI-constrained Multi-Armed bandit and propose an efficient learning algorithm, named O-SRS. Furthermore, we theoretically show that O-SRS achieves$O(\sqrt{T \log T})$regret while meeting the AoI requirement of each source. Finally, comprehensive simulations are conducted to demonstrate our theoretical results.

Unsupervised cross-domain image retrieval (UCDIR) aims to retrieve images across different domains without the guidance of labels. However, existing UCDIR methods assume that data can be shared across domains in plaintext, which is often impractical due to strict data privacy protection policies. In this paper, we propose ShieldIR, a novel privacy-preserving unsupervised cross-domain image retrieval framework that enhances the retrieval performance across two domains while safeguarding data privacy. ShieldIR unifies intra-domain and cross-domain representation learning through a Dual Protection Transformation (DPT) module, which introduces a structured feature space via orthogonal projection and ensures data privacy by adding calibrated differential privacy noise. This transformation allows ShieldIR to preserve semantic structure while formally protecting private data. For intra-domain representation learning, ShieldIR enhances discriminability by using DPT to map prototypes into an independent feature space and subsequently aligning the resulting dual-protected prototypes with instance-level features. For cross-domain alignment, ShieldIR maps intra-domain features and cross-domain prototypes into a shared structured space using DPT, achieving semantic alignment under privacy constraints. Extensive experiments on real-world datasets demonstrate that our ShieldIR outperforms state-of-the-art methods while effectively protecting data privacy.

Retrieval-augmented generation (RAG) paradigms can integrate external knowledge to enhance and validate the output of Large Language Models (LLMs) thereby mitigating generative hallucinations and broadening the model's knowledge scope. Despite advancements, existing RAG methods still suffer from uncertainty of prediction during the multi-round retrieval-generation process, and a lack of the ability to balance the adequacy and redundancy of retrieved information. To address these challenges, we propose CIRAG, an approach that combines the RAG process with collective intelligence. Inspired by the crowd of wisdom, CIRAG simulates individual independent decision-making and information aggregation within a crowd. Specifically, CIRAG first enhances retrieval diversity by expanding queries based on extracted entities, then combines frequency-based and semantic-based reranking to form a multi granularity fusion reranking thereby assessing better relevance, and integrate multiple information sources for accurate content generation. By undertaking these steps in an integrated manner, CIRAG enables the model to acquire comprehensive and non-redundant information for generating responses. We conduct extensive experiments with HotPotQA and 2WikiMultihopQA datasets, popular benchmark for retrieval-based, multi-step question-answering. Experimental results show that our approach surpasses existing advanced RAG framework while providing high portability in query expansion as well as strong comprehensiveness exhibited in the collective intelligence.

Searchable Symmetric Encryption (SSE) enables clients to make confidential queries over encrypted data while revealing some formally-defined leakage profiles. Despite the promising performance and application prospects of SSE, the recent leakage-abuse attacks show that a passive adversary can recover queries by exploiting patterns about data disclosed from leakage profiles. Among those attacks, the size pattern is a frequently exploited leakage. Although several countermeasures have been proposed, they can provide neither sufficient protection to mitigate size pattern leakage, nor sufficient scalability for large-scale databases. To address those challenges, we present an SGX-based size-pattern mitigation SSE schemeSMSSEwith two tailored response padding approaches and an I/O efficient disk-based index construction. In addition, we evaluate the size pattern leakage after padding through conditional entropy and differential privacy. Furthermore, we demonstrate the scalability robustness ofSMSSEon different databases by theoretically deducing the approximate boundary of index reading efficiency under a reasonable query distribution. Experiment results on representative real-world datasets show thatSMSSEcan provide high utility and strong protection against newly size pattern-based leakage-abuse attacks.

The design of effective online caching policies is an increasingly important problem for content distribution networks, online recommender systems, and edge computing services, etc. Exiting literature usually tackles this problem through the lens of optimistic online learning and aims to achieve sublinear regret. In this paper, we focus on a non-trivial extension of classic online caching problem inspired by operational requirements of real-world systems including switching costs and long-term constraints. To tackle the challenges of switching costs and operational long-term constraints in the online caching, we introduce the Block-structured Follow-the-Regularized-Leader (B-FTRL) caching policy. Our approach incorporates a block structure that divides time into blocks to minimize caching switching costs. The theoretical analysis shows that B-FTRL achieves a utility regret bound of $O\left( {{T^{\frac{{2a - b + 1}}{{1 + a}}}} + {T^{\frac{b}{{1 + a}}}}} \right)$ and switching costs bound of $O\left( {{T^{\frac{1}{{1 + a}}}}} \right)$, where a and b are tunable algorithm parameters. By carefully selecting the values of a and b, we are able to limit the total regret to O(T2/3) while satisfying the operational long-term constraints in expectation. Additionally, we provide high-probability constraint violation bounds of $O\left( {\sqrt T } \right)$. The performance of the proposed algorithm is evaluated with detailed trace-driven numerical tests.

Privacy-preserving content-based image retrieval aims to safeguard the security of outsourced private images while maintaining their searchability. However, existing schemes encounter challenges in striking a balance between security, accuracy, and efficiency, as well as difficulties in scaling to large-scale image retrieval in multi-user settings. In this paper, we propose a novel Efficient Large-Scale Encrypted Image Retrieval (ELSEIR) framework for outsourced data sharing. We first utilize a deep hashing model for image feature extraction. Building upon this, we design an irreversible random hash code generation method that incorporates permutation keys for personalized access and integrates differential privacy to further enhance data security. In our multi-user implementation, we distribute the switch keys to the cloud to standardize each key, enabling the accurate search. In addition, we have theoretically proven that our ELSEIR guarantees both outsourced data security and query user privacy. Extensive experiments on real-world datasets demonstrate that our ELSEIR yields comparable accuracy to the unprotected baseline while outperforming existing methods in terms of both retrieval accuracy and efficiency.

Signed graphs are prevalent data structures containing both positive and negative links. Recently, the fundamental network analysis task on signed graphs, namely link sign prediction, has received careful attention. Existing methods learn two target node representations independently, and the sign between these two nodes is predicted based on similarity. However, such a paradigm is node-centric that cannot distinguish node pairs with distinct contexts, thus lowering the prediction performance. Learning pair-centric representation is therefore a rewarding way to be aware of differences between pairs. There is no study yet on how to build such an appropriate representation that can effectively infer the sign between the target node pair. In this paper, we provide a new perspective to conduct link sign prediction within the paradigm of subgraph classification and propose a novel Subgraph-based link Sign Prediction (SSP) model. Technically, SSP uses importance-based sampling to extract an informative subgraph around each target node pair. For each subgraph, an innovative node labeling scheme is designed to encode its structural and signed information for representation learning. To further utilize the subgraph representation for imbalanced sign classification, SSP employs self-pruning contrastive learning to gain balanced representations. Extensive experiments on real-world datasets demonstrate that SSP consistently and significantly outperforms all the state-of-the-art baselines.

Recently, sequence generation methods are widely used in complex named entity recognition. By selecting high-related tokens to generate complex named entities, these methods obtain several achievements. However, due to lack of guidance in learning output format and ignoring labels in obtaining features, sequence generation methods suffer invalid output and inaccurate recognition. To solve that, we propose an Enhancing Entity Representation method with Prompt Guidance (ERPG). Specifically, in order to reduce invalid output, we design the candidate entity generation module that generate candidate entities and their labels as expected. Besides, to accurately recognize candidate entities, we propose candidate entity refine module, which obtain distinguishable candidate entity representations and filter them accurately. Based on that, our method finally outperforms baselines by 1.20, 1.62 and 0.69 F1 scores in ACE2004, GENIA and CADEC corpora, which proves the effectiveness in complex named entity recognition.

Relation extraction is the task of extracting relational facts between entities from plain text. When the extraction scope is extended to the document level, entities may exist in dif-ferent sentences. This requires the model to consider the in-teraction between multiple sentences comprehensively. Thus, document-level relation extraction becomes especially chal-lenging. Most existing models adopt entity representation learning to tackle this challenge. Nevertheless, they gener-ally perform representations of individual entities rather than directly modeling the dependencies between entities reflect target relation, which may cause wrong relation reasoning. To solve this issue, we propose a novel dual reasoning-based pairwise representation network. Specifically, we perform the semantic and syntactic-based reasoning to model the associ-ation between entities, and towards the capture of relational information in the document through an unique contextual se-lection for each entity pair. Experiments on three benchmark datasets demonstrate the inspiring performance improvement over state-of-the-art relation extraction models.

Sequential recommendations can dynamically model user interests, which has great value since users' interests may change rapidly with time. Traditional sequential recommendation methods assume that the user behaviors are rigidly ordered and sequentially dependent. However, some user behaviors have flexible orders, meaning the behaviors may occur in any order and are not sequentially dependent. Therefore, traditional methods may capture inaccurate user interests based on wrong dependencies. Motivated by this, several methods identify flexible orders by continuity or similarity. However, these methods fail to comprehensively understand the nature of flexible orders since continuity or similarity do not determine order flexibilities. Therefore, these methods may misidentify flexible orders, leading to inappropriate recommendations. To address these issues, we propose a Flexible Order aware Sequential Recommendation (FOSR) method to identify flexible orders comprehensively. We argue that orders' flexibilities are highly related to the frequencies of item pair co-occurrences. In light of this, FOSR employs a probabilistic based flexible order evaluation module to simulate item pair frequencies and infer accurate order flexibilities. The frequency labeling module extracts labels from the real item pair frequencies to guide the order flexibility measurement. Given the measured order flexibilities, we develop a flexible order aware self-attention module to model dependencies from flexible orders comprehensively and learn dynamic user interests effectively. Extensive experiments on four benchmark datasets show that our model outperforms various state-of-the-art sequential recommendation methods.

Community search, also known as local community detection, is designed to obtain the community structure containing the given query nodes. At present, community search technology has been widely studied, but most of them ignore the effect of node attributes. Some studies propose attributed community search methods and usually adopt a two-phase approach. Generally, the subgraph with structure cohesiveness is obtained first and then the subgraph with keyword cohesiveness is returned as the final result on this basis. In practice, approaches of this kind may lead to incomplete or incorrect results, because information from one single aspect cannot accurately describe the feature of nodes. That is, topology information and attribute information of the network should be complementary. In this paper, we propose an attributed community model called RCS(Representation-based Community Search). Firstly we apply random-walk based representation learning technology to combine topology information and attribute information of nodes. Secondly we propose a new community model to formulate the problem that what is a good community based on the representation information. Then a series of algorithms are designed to fetch a cohesive community whose vertices are tightly connected and closely related to query nodes. Finally the effectiveness and efficiency of RCS are verified by extensive experiments on several real-world networks with ground-truth cnmmunities.