Retrieval-Augmented Generation (RAG) mitigates hallucinations in large language models (LLMs) by incorporating external knowledge in open-domain question answering. However, for complex multi-hop questions, mainstream iterative retrieval methods struggle with conflicts between internal and retrieved knowledge and noisy external documents. Achieving flexible and effective integration of internal and external knowledge continues to be a research challenge. In this paper, we propose a dynamic knowledge fusion-based RAG framework (DKF-RAG), which can adaptively adjust the knowledge fusion strategy during the iterative reasoning process. Specifically, DKF-RAG iteratively decomposes complex multi-hop questions into simpler sub-questions, forming a reasoning chain. At each iteration, it assesses the relevance and consistency of external knowledge in real time, based on which adaptive fusion actions are triggered to integrate internal and external information. We conduct experiments on four multi-hop QA datasets and compare DKF-RAG against multiple iterative retrieval-augmented baselines, demonstrating its superiority and effectiveness.

Text style transfer (TST) is an important task in natural language generation, which aims to transfer the text style (e.g., sentiment) while keeping its semantic information.Due to the absence of parallel datasets for supervision, most existing studies have been conducted in an unsupervised manner, where the generated sentences often suffer from high semantic divergence and thus low semantic preservation.In this paper, we propose a novel disentanglementbased framework for TST named DisenTrans, where disentanglement means that we separate the attribute and content components in the natural language corpus and consider this task from these two perspectives.Concretely, we first create a disentangled Chain-of-Thought prompting procedure to synthesize parallel data and corresponding attribute components for supervision.Then we develop a disentanglement learning method with synthetic data, where two losses are designed to enhance the focus on attribute properties and constrain the semantic space, thereby benefiting style control and semantic preservation respectively.Instructed by the disentanglement concept, our framework creates valuable supervised information and utilizes it effectively in TST tasks.Extensive experiments on mainstream datasets present that our framework achieves significant performance with great sample efficiency.

With the rapid growth of wireless sensor networks, secure data transmission, storage, and distribution in such networks has become an urgent demand. To defend against security risks such as data leakage, key compromise, and unauthorized misuse of data simultaneously, we propose a novel obfuscatable ciphertext-policy attribute-based re-encryption scheme with a specially designed obfuscator. The proposed scheme leverages program obfuscation to transform the re-encryption program codes into an unintelligible form and embed the private keys into the obfuscated implementation. Consequently, the proposed scheme protects data confidentiality and keeps the secrecy of the private key while providing fine-grained access control. Formal proofs for the security of the proposed re-encryption scheme and the obfuscator are provided. Extensive experiments have been conducted on representative platforms, including cloud servers, workstations, and embedded devices, to evaluate the computational efficiency and energy consumption of the scheme. Experimental results indicate that the scheme achieves high efficiency on various platforms and economical energy consumption on typical embedded devices with constrained resources.

Text style transfer (TST) is an important task in natural language generation, which aims to alter the stylistic attributes (e.g., sentiment) of a sentence and keep its semantic meaning unchanged.Most existing studies mainly focus on the transformation between styles, yet ignore that this transformation can be actually carried out via different hidden transfer patterns.To address this problem, we propose a novel approach, contrastive transfer pattern mining (CTPM), which automatically mines and utilizes inherent latent transfer patterns to improve the performance of TST.Specifically, we design an adaptive clustering module to automatically discover hidden transfer patterns from the data, and introduce contrastive learning based on the discovered patterns to obtain more accurate sentence representations, and thereby benefit the TST task.To the best of our knowledge, this is the first work that proposes the concept of transfer patterns in TST, and our approach can be applied in a plug-andplay manner to enhance other TST methods to further improve their performance.Extensive experiments on benchmark datasets verify the effectiveness and generality of our approach.1

Controllable text generation (CTG) aims to generate text with desired attributes, and decodingtime-based methods have shown promising performance on this task.However, in this paper, we identify the phenomenon of Attribute Collapse for the first time.It causes the fluency of generated text to rapidly decrease when the control strength exceeds a critical value, rendering the text completely unusable.This limitation hinders the effectiveness of decoding methods in achieving high levels of controllability.To address this problem, we propose a novel lightweight decoding framework named Air-Decoding.Its main idea is reconstructing the attribute distributions to balance the weights between attribute words and nonattribute words to generate more fluent text.Specifically, we train prefixes by prefix-tuning to obtain attribute distributions.Then we design a novel attribute distribution reconstruction method to balance the obtained distributions and use the reconstructed distributions to guide language models for generation, effectively avoiding the issue of Attribute Collapse.Experiments on multiple CTG tasks prove that our method achieves a new state-of-the-art control performance 1 .

Mobile agents are useful in collaborative business systems due to their mobility and autonomy, which can roam over the Internet to purchase goods and services on behalf of their owners. However, given attacks from a malicious host, it is a challenge to securely sign a contract on behalf of the owner (the original signer). In this paper, we propose an intrusion-resilient undetachable digital signature (IR-UDS) approach to mitigate the security risk of signing key leakage on the signer's host, base device, and potentially malicious remote hosts, as well as the risk of misusing the signing algorithm on remote hosts. An attacker will be unable to forge the past and future signatures as long as the base device is secure, even if the current signing key of the original signer has been gained. When the base device is compromised, although the future signatures could be forged, all past signatures remain secure. Furthermore, the encrypted signing function has been combined with the original signer's requirement to prevent the misuse of signing algorithm and the exposure of original signing key on malicious hosts. Security analysis has indicated that our scheme can defeat a variety of attacks, and experimental evaluations have demonstrated the good performance of the scheme.

Signcryption is a cryptographic technique for simultaneously performing both digital signature and data encryption. It is effective for protecting the confidentiality and unforgeability of communications in Internet of Things (IoT) systems, especially when a number of generated ciphertexts can be aggregated into a compact form. However, device capture attacks are commonly threatening the implementations of signcryption on unattended devices by enabling an attacker to extract the cryptographic key from a captured device. Motivated by this issue, we propose a novel and specialized obfuscatable aggregatable signcryption scheme (OASC) together with an obfuscator for the signcryption algorithm, which has been designed by taking into account that the computational and communication costs should be sufficiently small (light-weighted) to fit applications in resource-constrained embedded devices. The proposed obfuscator can protect signcryption programs from key-extraction attacks by transforming the programs into unintelligible obfuscated programs. To the best of our knowledge, this is the first OASC in the community. The scheme's security features with respect to obfuscation, confidentiality, and unforgeability have been theoretically proved. Moreover, in comparison with other (nonobfuscatable) aggregatable signcryption schemes, the scheme's computational efficiency is positioned at a medium level while the communication cost is also relatively small, with extra unique security features benefiting from obfuscation. Experiments on different devices indicated that the proposed scheme performs reasonably well as expected. The scheme is widely applicable for various scenarios of IoT, where information is sent from unattended leaf nodes to a sink point.