Training Large Vision-Language Models (LVLMs) is costly and resource-intensive, making them valuable assets.To prevent malicious users from unauthorized commercialization of these artificial intelligence assets through finetuning and black-box deployment, model fingerprinting techniques aimed at verifying the ownership of LVLMs are receiving widespread attention.Existing fingerprinting techniques rely on adversarial attacks or backdoor attacks to construct trigger images for specific outputs, attributing model ownership by comparing whether the output of trigger images on suspected models matches the predetermined output.However, these methods depend on fixedform triggers as explicit model fingerprints, which have limitations in terms of stealthiness and robustness.Inspired by unlearning research, we propose Unlearning-based Multimodal Memorization Fingerprint (UMMF).UMMF strengthens the overfitting characteristics of training samples by unlearning neighboring samples of the training samples, thereby introducing detectable regions of poor generalization in the data manifold.Compared with previous methods, our approach leverages the differences in memorization strength of LVLMs on neighboring samples as implicit model fingerprints, rather than relying on specific inputoutput pairs as explicit triggers.This endows it with stronger stealthiness, robustness, and adaptability.To simulate real application scenarios, we conduct extensive experiments using multiple strategies and different datasets, further demonstrating its superiority in protecting LVLM ownership.

In the era of social media, the proliferation of fake news has created an urgent need for more effective detection methods, particularly for multimodal content. The task of identifying fake news is highly challenging, as it requires broad background knowledge and understanding across various domains. Existing detection methods primarily rely on neural networks to learn latent feature representations, resulting in black-box classifications with limited real-world understanding. To address these limitations, we propose a novel approach that leverages Multimodal Large Language Models (MLLMs) for fake news detection. Our method introduces adversarial reasoning through debates from opposing perspectives. By harnessing the powerful capabilities of MLLMs in text generation and cross-modal reasoning, we guide these models to engage in multimodal debates, generating adversarial arguments based on contradictory evidence from both sides of the issue. We then utilize these arguments to learn reasonable thinking patterns, enabling better multimodal fusion and fine-tuning. This process effectively positions our model as a debate referee for adversarial inference. Extensive experiments conducted on four fake news detection datasets demonstrate that our proposed method significantly outperforms state-of-the-art approaches.

The rapid development of social media has led to a surge of eye-catching fake news on the Internet, with multimodal news comprising both images and text being particularly prevalent. To address the challenges of Multimodal Fake News Detection (MFND), numerous supervised task-specific Multimodal Small Language Models (MSLMs) have been developed. However, these models lack the breadth of knowledge and the depth of language understanding, which results in unsatisfactory adaptability, generalization, and explainability performance. To address these issues, we attempt to introduce Large Vision-Language Models (LVLMs), aiming to leverage the common sense understanding and logical reasoning abilities of LVLMs for the MFND task. We observed that LVLMs can generate reasonable analyses of news content from specific angles. However, when it comes to synthesizing these analyses for final judgment, their performance declines significantly, failing to meet the accuracy benchmarks set by existing MSLMs detection models. This reflects the need for a more effective way for LVLMs, which have not undergone task-specific training, to utilize their knowledge and capabilities. Based on these findings, we propose the Explainable Adaptive Rationale-Augmented Multimodal (EARAM) framework, which adaptively uses MSLMs to extract useful rationales from the multi-perspective analyses of LVLMs. After making judgments based on these rationales, EARAM then assists LVLMs in generating more reliable explanations. Extensive experiments demonstrate that our model not only achieves state-of-the-art results on widely used datasets but also significantly outperforms other models in terms of generalization and explainability.

Public-key Encryption with Keyword Search (PEKS) is a promising cryptographic mechanism that enables a semi-trusted cloud server to perform (on-demand) keyword searches over encrypted data for data users. Existing PEKS schemes are limited to precise or fuzzy keyword searches, creating a gap given the widespread use of wildcards for rapid searches in real-world applications. To address this issue, several wildcard keyword search schemes have been proposed to support wildcard searches in the public-key setting. However, these schemes suffer from inefficiency and/or inflexibility. Worse yet, they are all vulnerable to (insider) keyword guessing attacks (KGA), which is highly effective when the keyword space is polynomial in size. To address these vulnerabilities, this paper first proposes a new wildcard keyword search scheme called Public-key Encryption with Wildcard Search (PEWS), which is built based on the standard Decisional Diffie-Hellman (DDH) assumption. The complexity of all algorithms in PEWS increases linearly with the keyword length, while remaining almost constant or even decreasing linearly with the number of wildcards. To resist against (insider) KGA, we further extend PEWS into the first Public-key Authenticated Encryption with Wildcard Search (PAEWS) scheme. Our PEWS and PAEWS schemes are highly flexible, supporting searches for any number of wildcards positioned anywhere within the keyword. We conduct a comprehensive performance evaluation of our PEWS and PAEWS, while also comparing PEWS with the state-of-the-art scheme in the public-key setting. The experimental results demonstrate that both PEWS and PAEWS are efficient and practical, and the experimental comparisons illustrate that PEWS achieves approximately$2 \times $faster computation and reduces communication by at least 50%.

Proxy re-encryption (PRE), as a promising cryptographic primitive for secure data sharing in clouds, has been widely studied for decades. PRE allows the proxies to use the re-encryption keys to convert ciphertexts computed under the delegator’s public key into ones that can be decrypted using the delegatees’ secret keys, without knowing anything about the underlying plaintext. This delegable property of decryption rights enables flexible cloud data sharing, but it raises an important issue: if some proxies reveal their re-encryption keys, or collude with some delegatees to create a pirate decoder, then anyone who gains access to the pirate decoder can decrypt all ciphertexts computed under the delegator’s public key without the delegator’s permission. This paper opens up a potentially new avenue of research to address the above (re-encryption) key abuse problem by proposing the first public trace-and-revoke PRE system, where the malicious delegatees and proxies involved in the generation of a pirate decoder can be identified by anyone who gains access to the pirate decoder, and their decryption capabilities can subsequently be revoked by the content distributor. Our construction is multi-hop, supports user revocation and public (black-box) traceability, and achieves significant efficiency advantages over previous constructions. Technically, our construction is a generic transformation from inner-product functional PRE (IPFPRE) that we introduce to trace-and-revoke PRE. In addition, we instantiate our generic construction of trace-and-revoke PRE from the Learning with Errors (LWE) assumption, which was widely believed to be quantum-resistant. This is achieved by proposing the first LWE-based IPFPRE scheme, which may be of independent interest. Finally, we conduct a comprehensive performance evaluation of our LWE-based trace-and-revoke PRE scheme, and the experimental results show that the proposed LWE-based trace-and-revoke PRE scheme is practical and outperforms current state-of-the-art traceable PRE schemes.