Retrieval-Augmented Generation (RAG) has been proposed to mitigate hallucinations in large language models (LLMs), where generated outputs may be factually incorrect.However, existing RAG approaches predominantly rely on vector similarity for retrieval, which is prone to semantic noise and fails to ensure that generated responses fully satisfy the complex conditions specified by factual queries.To address this challenge, we introduce a novel research problem, named EXACT RETRIEVAL PROBLEM (ERP).To the best of our knowledge, this is the first problem formulation that explicitly incorporates structural information into RAG for factual questions to satisfy all query conditions.For this novel problem, we propose STRUCTURE GUIDED RETRIEVAL-AUGMENTED GENERATION (SG-RAG) 1 , which models the retrieval process as an embedding-based subgraph matching task, and uses the retrieved topological structures to guide the LLM to generate answers that meet all specified query conditions.To facilitate evaluation of ERP, we construct and publicly release EXACT RETRIEVAL QUESTION ANSWERING (ERQA) 2 , a large-scale dataset comprising 120,000 fact-oriented QA pairs, each involving complex conditions, spanning 20 diverse domains.The experimental results demonstrate that SG-RAG significantly outperforms strong baselines on ERQA, delivering absolute gains of 20.68-50.88percentage points, corresponding to 34%-450% relative improvements across metrics, while maintaining reasonable computational overhead.

With the rapid development of artificial intelligence technology, large language models (LLMs) have shown tremendous potential in multiple fields. This article aims to explore the use of Low Rank Adaptation (LoRA) technology to fine tune the Llama3 model with increased fishery datasets and apply it to the field of fisheries to improve the intelligence level of fisheries management. The Llama3 model, developed by Meta, is renowned for its robust language understanding and generation capabilities. However, fine-tuning this model with all parameters is highly resource-intensive and time-consuming. To address these challenges, this study employs LoRA technology, which leverages low-rank decomposition to efficiently fine-tune the Llama3 model. By doing so, we are able to adapt the model to meet the specific requirements of the fishery domain while significantly reducing computational costs and training time. In the field of fisheries, the Llama3 model fine-tuned by LoRA is used to process and analyze textual data related to fisheries, such as water quality and environmental monitoring, fish behavior analysis, fish farming techniques, and interpretation of fisheries policies. By fine-tuning, the model can more accurately understand the professional terminology and contextual information in the field of fisheries, thereby improving the accuracy and efficiency of information processing. This article introduces the basic principles of the Llama3 model and LoRA fine-tuning technology, and elaborates on the specific implementation method of applying LoRA technology to the Llama3 model. A series of experiments were designed to evaluate the application effect of the Llama3 model fine-tuned by LoRA in the field of fisheries. The experimental results indicate that the fine-tuned model demonstrates notable enhancements in performance across various tasks, including fishery text classification, information extraction, and question answering systems. Additionally, the study explores the strengths and challenges of employing LoRA fine-tuning technology within the fishery domain. In terms of challenges, further research is needed on how to optimize the model structure, expand the fishery datasets, improve the model generalization ability, and cope with the complex and changing environment in the fishery field.

ABSTRACT With fast development of cloud computing, more and more sensitive data are stored in distributed storage systems. It is mainly suffered from the following two threats: (i) data at rest are stolen; and (ii) data in transmission are intercepted. The threats may lead to serious problems. For example, personal private data eavesdropping may lead to legal problems and credit crisis, while company information leaking may cause huge economic loss. To tackle such security threat in cloud distributed storage system, we propose a novel scheme to produce dynamic size of shares with multiple iterations of secret sharing scheme. A file is partitioned into multiple file shares, and one of them is taken as a new file to be further partitioned into smaller size with low additional computational cost and data expansion. The small‐size file share is stored at user terminal as a necessary component to recover original file. It can be used in scenarios that user terminal, such as smart phone, is one of nodes in distributed storage systems. In this case, attackers cannot recover information through eavesdropping shares transmitted in network. Storage service providers cannot also obtain useful information through file shares stored online without the participant of the small‐size file share at user terminal. Therefore, sensitive data can be protected confidentially. Copyright © 2013 John Wiley & Sons, Ltd.

Most of secret sharing schemes have to be computed in a Galois field, such as Shamir's scheme, which have relatively heavy computational cost. Kurihara et al. recently proposed a fast secret sharing scheme using only Exclusive-OR(XOR) operations to make shares and recover the secret. Their proposed scheme was shown to be hundreds of times faster than Shamir's (in GF(q=264)) in terms of both distribution and recovery with a 4.5 MB secret when k=3 and n=11. However, some steps in their scheme still need to be improved. Their security proofs were too complex and difficult to be understood and verified intuitively. In this paper, we present a conciser, cleaner, faster scheme which is also based on XOR. Moreover, we give two geometric explanations of making shares in both our and Kurihara's schemes respectively, which would help to easier and further understand how the shares are made in the two schemes.