Automatically translated preprints from the Chinese arXiv (ChinaXiv).
Showing 50 of 110 papers in AI / CS
Advances in technology and the development of the Internet have enabled the storage and distribution of large-scale unstructured data (such as audio, video, and natural language text). However, any such storage and distribution incurs certain costs, naturally prompting considerations for the efficie…
We point out that neural networks are not black boxes, and their generalization stems from the ability to dynamically map a dataset to the local extrema of the model function. We further prove that the number of local extrema in a neural network is positively correlated with the number of its parame…
Investigating the influencing factors of emergency information acquisition among rural left-behind children during major public health events can provide valuable recommendations for improving their emergency information acquisition capabilities. This study conducted interviews with 34 rural left-be…
The phenomenon of citizens sharing unverified information on social media can, to a certain extent, reflect the level of information literacy education in a region. This study aims to investigate the attitudes and behaviors of the general public in Taiwan, China toward unverified information on soci…
Micro-expressions refer to brief facial actions that individuals unconsciously display when attempting to suppress genuine emotions. Owing to their non-invasive nature, they possess significant application value in domains such as national security and public security. To address challenges includin…
We present a new method for structural sequence analysis grounded in Algorithmic Information Theory (AIT). At its core is the Ladderpath approach, which extracts nested and hierarchical relationships among repeated substructures in linguistic sequences---an instantiation of AIT’s principle of descri…
Mathematical reasoning presents a significant challenge for Large Language Models (LLMs) as it requires ensuring the correctness of each reasoning step. Researchers have been strengthening the mathematical reasoning abilities of LLMs through supervised fine-tuning, but due to the inability to suppre…
[Purpose/Significance] This study explores the adaptability and evolution path of Agent technology in scientific and technological intelligence work, addressing the increasingly growing demands for intelligence and cognition in intelligence operations. [Method/Process] Based on the functional evolut…
Natural language is considered closely intertwined with human cognition, with linguistic structures posited to offer profound insights into the cognitive system. However, as a coding system, natural language encodes diverse objects into unified forms; its prominent formal features capture people’s a…
With the rapid penetration of artificial intelligence across multiple industries and scenarios, integrating the language understanding advantages of general large models with industry-specific knowledge bases in complex real-world applications has become a key challenge in constructing next-generati…
Objective: With the development of internet big data and machine learning methods, an increasing number of studies combine text analysis and machine learning to predict satisfaction. In research on building life satisfaction prediction models, to address the challenge of obtaining large amounts of e…
This paper analyzes the major challenges currently facing Large Language Models (LLMs) and proposes concrete solutions. It identifies that the representation and computation of probabilistic conceptual structure models is crucial, provides a brief overview of the related technology—Deep Semantic Mod…
OpenAI's Generative Pre-trained Transformer 4 (GPT-4) is a powerful large language model with a certain degree of intelligence in understanding and generating coherent text. We are exploring whether GPT-4 is capable of acting as a die, i.e. generating random numbers. We show that GPT-4 does not appe…
To promote critical improvement of systems related to "external regulations for research integrity" both within and beyond the scientific community, this study examines the practical drivers and theoretical logic underlying the institutional practice of "random inspection of national science and tec…
[Objective] Analyze the definition and properties of information capacity based on the information sextuple model, propose formulas relating information to matter, energy, and time, and provide theoretical reference for information science, particularly for applications in quantum information techno…
Until now, image quality assessment has not addressed color issues. The literature on image quality assessment primarily evaluates the degree of degradation (deterioration) of image quality during image processing procedures such as compression and transmission. A planar image is essentially a two-d…
[Purpose] Large scientific facilities play a crucial role in promoting the output of major scientific and technological achievements, and their clustered development and collaborative innovation effects continuously lead regional industrial innovation and development. To enable large facilities to b…
To address issues such as low registration efficiency and large errors in the registration of two partially overlapping point clouds, we propose a point cloud registration algorithm based on mixed features sampled from the overlapping region. First, we predict the overlap score for each point throug…
Bone age assessment is a commonly used method for detecting endocrine and growth abnormalities in children; however, low-quality hand X-ray images in deep learning methods reduce the final assessment accuracy. To address this issue, we propose an alignment network that increases the region of intere…
Employing reinforcement learning to address robot manipulation problems offers numerous advantages; however, traditional reinforcement learning algorithms encounter challenges associated with sparse rewards, and the policies obtained are difficult to directly deploy in real-world environments. To en…
To address the problems of ambiguous boundaries between high-value and low-value parts and unreasonable disassembly depth of waste products, a multi-objective disassembly scheme decision-making method for products considering functional degradation of parts is proposed. The influence of specific par…
In recent years, the application of unmanned aerial vehicles (UAVs) has become increasingly widespread, and the coordination of multiple UAVs to accomplish tasks has substantially improved operational efficiency. Motivated by this phenomenon, numerous scholars have devoted themselves to research on …
To resist quantum algorithm attacks and address the vulnerability where malicious signers exploit the complete anonymity of ring signature technology to output multiple signatures for double-spending attacks, while simultaneously solving the problem of unnecessary system overhead waste, a novel latt…
To address the issues of high latency in existing low-orbit satellite network authentication schemes that utilize centralized authentication methods and significant computational overhead arising from complex bilinear mappings, we introduce a certificateless authentication model and propose an effic…
Traditional ciphertext retrieval schemes in cloud environments generate document vectors and query vectors based on statistical models, without considering the deep semantic information of documents and requests. This paper proposes a deep semantic ciphertext retrieval model based on a hybrid cloud …
To address the problems of low delivery timeliness and poor customer value in the logistics industry resulting from increasing vehicle ownership and escalating traffic congestion, this study proposes a time-varying routing optimization method for truck-drone collaborative delivery, comprehensively c…
Synthetic Minority Over-sampling Technique (SMOTE) is one of the effective methods for addressing class imbalance problems. However, SMOTE's linear interpolation mechanism restricts synthetic samples to the line connecting original samples, resulting in a lack of diversity in new samples, and this l…
To address the fresh food closed-loop supply chain network design problem, a robust optimization model for fresh food closed-loop supply chain networks is established to resolve uncertainties in the supply chain network. First, for a fresh food supply chain network structure encompassing 5 nodes, a …
To enhance the solution efficiency for the Multi-Depot Vehicle Routing Problem (MDVRP), we propose an end-to-end deep reinforcement learning framework. First, MDVRP is formulated as a Markov Decision Process (MDP), encompassing definitions of state, action, and reward. Additionally, we propose an im…
Text representation needs to address the ambiguity of textual terms and accurately define their semantic features within specific contextual environments. To tackle the problems of word polysemy and contextual characteristics, we propose an SCDVAB model for text semantic disambiguation. The main inn…
To address the issues that the Raft algorithm cannot resist attacks from Byzantine nodes and that logs are susceptible to tampering and forgery, we design an RB-Raft (Resist Byzantine-Raft) algorithm that resists Byzantine nodes. First, we employ a hash chain approach to perform iterative hashing on…
To address the issues of time-consuming training, large memory consumption, and difficulty in model updating associated with traditional offline hashing algorithms, as well as the phenomenon of substantial label loss in real-world image datasets, we propose an online hashing algorithm for balanced l…
To investigate the effects of hazard factors such as temperature, smoke, and CO concentration on evacuation in fire scenarios, a fire evacuation model based on dynamic coupling of FDS and cellular automata is established. The model establishes a one-to-one correspondence between FDS grids and cellul…
To achieve accurate classification of breast pathology WSI images, we propose a gated convolutional neural network classification method based on hybrid connections. A hybrid module incorporating local residual connections and global dense connections is constructed, with squeeze-and-excitation gate…
Complex event processing is a technique for analyzing event streams in dynamic environments. Typically implemented based on finite state automata, complex event processing techniques generate a large number of overlapping partial matches on the event stream during the matching process, requiring the…
A bi-level programming model is established to address the actual battery swapping characteristics of automated guided vehicles (AGVs) in automated container terminals, aiming to reduce the total task completion time and total battery swapping time of AGVs while rationally planning the number of bat…
To address the issues of slow convergence and susceptibility to local optima in existing ant colony optimization-based algorithms for solving distributed constraint optimization problems, this paper proposes a Random disturbance based multi-population ant colony algorithm to solve distributed constr…
To address the suboptimal segmentation performance of existing visible-infrared (RGB-T) image semantic segmentation models, this paper proposes a nested segmentation network based on deep difference feature complementary fusion. Specifically, the encoder and decoder components are interconnected via…
Mobile crowdsensing data contains image and spatio-temporal contextual information that can be utilized for detecting changes in street view images; however, such data is typically low-quality and non-standard. To accurately detect changes in street scenes, this paper primarily addresses the data qu…
To address the deficiencies of the School-Based Optimization (SBO) algorithm, including poor search performance and susceptibility to local optima, this paper proposes an SBO algorithm integrated with educational psychology (SBO based on Educational Psychology, SBO-EP). During the teaching phase, th…
To improve the accuracy of overlapping community detection, we propose CLPANNI (Cycle Label Propagation Algorithm with Neighbor Node Influence), an optimized algorithm based on cycle structures derived from LPANNI. The algorithm extracts minimal cycle information of nodes, measures node importance u…
Social media account credibility assessment is an important component in ensuring the healthy development of online social ecosystems. To address issues such as multi-dimensional credibility assessment indicators for social media accounts and diverse data information uncertainties, a credibility ass…
In fifth-generation mobile communications, network slicing is employed to provide an optimal network for diverse services. In the context of RAN slicing scenarios across multiple base stations, conventional resource allocation methods fail to meet slice requirements when the number of slices varies …
Link prediction is an important problem in the field of data mining. Similarity methods based on random walks generally assume that the probability of a walking particle transferring to neighboring nodes is equal, ignoring the influence of node degree values on the transfer probability. To address t…
Named Data Networking (NDN), as a novel Internet architecture, aims to address the ever-increasing data traffic. Based on its consumer-driven content retrieval model, NDN naturally supports consumer mobility. However, producer mobility remains a challenging problem, requiring additional mechanisms t…
Bullens et al. left an open problem in CSI-Fish, namely designing an identification protocol that allows the system challenge space to be #1;, rather than a small set #1;. This paper proposes a zero-knowledge proof scheme based on supersingular isogenies. The scheme treats the challenge C as an isog…
Existing mobile crowdsensing task recommendation methods share common drawbacks: on one hand, they fail to fully consider the impact of spatio-temporal information on worker preferences, resulting in low recommendation accuracy; on the other hand, they ignore the influence of task popularity on reco…
In the IoT service discovery process, users typically express their requirements through their own intentions, whereas service descriptions constitute explanations of service functionalities; consequently, mismatches between these two elements impact the accuracy of service discovery. Simultaneously…
Due to their limited battery life and computing power, autonomous vehicles struggle to meet the processing demands of latency-sensitive or compute-intensive tasks while ensuring driving range. To address this issue, this paper proposes an autonomous vehicle task offloading strategy based on deep Q-n…
Analyzing student evaluation data for teachers in teaching evaluation systems enables teachers to comprehend students' authentic attitudes toward instructors, summarize teaching experience, improve subsequent teaching methods, and enhance teaching quality. However, during teaching evaluations, issue…
