A5076107339- Network Security and Intrusion Detection
- Internet Traffic Analysis and Secure E-voting
- Postharvest Quality and Shelf Life Management
- Network Packet Processing and Optimization
- Plant Physiology and Cultivation Studies
- Gaussian Processes and Bayesian Inference
- Leaf Properties and Growth Measurement
- Spectroscopy and Chemometric Analyses
- Topic Modeling
- Advanced Graph Neural Networks
- Model Reduction and Neural Networks
- Bayesian Methods and Mixture Models
- Smart Agriculture and AI
- Neural Networks and Applications
- Data Stream Mining Techniques
- Complex Network Analysis Techniques
- FinTech, Crowdfunding, Digital Finance
- Impact of Light on Environment and Health
- Robotics and Sensor-Based Localization
- Transportation Planning and Optimization
- China's Ethnic Minorities and Relations
- Human Mobility and Location-Based Analysis
- Human Pose and Action Recognition
- Private Equity and Venture Capital
- IPv6, Mobility, Handover, Networks, Security
Shanxi Agricultural University
2023-2024
Chinese Academy of Sciences
2019-2023
Institute of Information Engineering
2019-2023
Beijing University of Chemical Technology
2021-2022
University of Chinese Academy of Sciences
2018-2021
Helsinki Institute for Information Technology
2020
Aalto University
2020
University of Technology
2019
Institute of Software
2018
Large language models (LLMs) have strong capabilities in solving diverse natural processing tasks. However, the safety and security issues of LLM systems become major obstacle to their widespread application. Many studies extensively investigated risks developed corresponding mitigation strategies. Leading-edge enterprises such as OpenAI, Google, Meta, Anthropic also made lots efforts on responsible LLMs. Therefore, there is a growing need organize existing establish comprehensive taxonomies...
Cities are typical dynamic complex systems that connect people and facilitate interactions. Revealing general collective patterns behind spatiotemporal interactions between residents is crucial for various urban studies, of which we still lacking a comprehensive understanding. Massive cellphone data enable us to construct interaction networks based on co-occurrence individuals. The rank-size distributions population locations in all unit time windows stable, although almost constantly moving...
To detect quickly and accurately “Yuluxiang” pear fruits in non-structural environments, a lightweight YOLO-GEW detection model is proposed to address issues such as similar fruit color leaves, bagging, complex environments. This improves upon YOLOv8s by using GhostNet its backbone for extracting features of the pears. Additionally, an EMA attention mechanism was added before fusing each feature neck section make focus more on target information fruits, thereby improving recognition ability...
Deep neural networks, such as Faster R-CNN, have been widely used in object detection.However, deep networks usually require a large-scale dataset to achieve desirable performance.For the specific application, UAV detection, training data is extremely limited practice.Since annotating plenty of images manually can be very resource intensive and time consuming, instead, we use PBRT render large number photorealistic high variation within reasonable time.Using ensures realism rendered images,...
The intelligent detection of young peaches is the main technology fruit-thinning robots, which crucial for enhancing peach fruit quality and reducing labor costs. This study presents lightweight YOLO-PEM model based on YOLOv8s to achieve high-precision automatic “Okubo” peaches. Firstly, C2f_P module was devised by partial convolution (PConv), replacing all C2f modules in model’s lightweight. Secondly, embedding efficient multi-scale attention (EMA) C2f_P_1 backbone network enhanced feature...
Evaluating image captions typically relies on reference captions, which are costly to obtain and exhibit significant diversity subjectivity. While reference-free evaluation metrics have been proposed, most focus cross-modal between images. Recent research has revealed that the modality gap generally exists in representation of contrastive learning-based multi-modal systems, undermining reliability cross-modality like CLIPScore. In this paper, we propose CAMScore, a cyclic automatic metric...
Cities are centers for the integration of capital and incubators invention, attracting venture (VC) is great importance cities to advance in innovative technology business models towards a sustainable prosperous future. Yet we still lack quantitative understanding relationship between urban characteristics VC activities. In this paper, find clear nonlinear scaling activities population Chinese cities. such systems, widely applied linear per capita indicators would be either biased larger or...
Global IPv6 scanning has always been a challenge for researchers because of the limited network speed and computational power. Target generation algorithms are recently proposed to overcome problem Internet assessments by predicting candidate set scan. However, custom address configuration emerges diverse addressing patterns discouraging algorithmic inference. Widespread alias could also mislead algorithm discover aliased regions rather than valid host targets. In this paper, we introduce...
Estimating global pairwise interaction effects, i.e., the difference between joint effect and sum of marginal effects two input features, with uncertainty properly quantified, is centrally important in science applications. We propose a non-parametric probabilistic method for detecting unknown form. First, relationship features output modelled using Bayesian neural network, capable representing complex interactions principled uncertainty. Second, their are estimated from trained model. For...
Unlike IPv4 addresses, which are typically masked by a NAT, IPv6 addresses could easily be correlated with user activity, endangering their privacy. Mitigations to address this privacy concern have been deployed, making existing approaches for address-to-user correlation unreliable. This work demonstrates that an adversary still correlate users accurately, even these protection mechanisms. To do this, we propose model - SiamHAN. The uses Siamese Heterogeneous Graph Attention Network measure...
Since the lack of IPv6 network development, China is currently accelerating deployment. In this scenario, traffic and structure show a huge shift. However, due to long-term prosperity, we are ignorant problems behind such outbreak performance improvement events in development some regions will still face similar challenges future. To contribute solving problem, paper, produce new measurement framework implement 5-month passive on during deployment China. We combine 6 global-scale datasets...
Abstract To address the challenges of high similarity in height between young fruits and leaves, small size fruits, dense distribution, severe occlusions, this paper proposes a lightweight YOLOv8s-P detection model for identification "Okubo" peaches complex environments. Firstly, C2f_Faster module is designed replaces all C2f modules YOLOv8s to realize lightweight. Secondly, Efficient Multi-Scale Attention Module(EMA) added inside enhance network's ability extract tiny features. Finally,...
Log analysis is crucial for ensuring the orderly and stable operation of information systems, particularly in field Artificial Intelligence IT Operations (AIOps). Large Language Models (LLMs) have demonstrated significant potential natural language processing tasks. In AIOps domain, they excel tasks such as anomaly detection, root cause faults, operations maintenance script generation, alert summarization. However, performance current LLMs log remains inadequately validated. To address this...
Fast IPv6 scanning is challenging in the field of network measurement as it requires exploring whole address space but limited by current computational power. Researchers propose to obtain possible active target candidate sets probe algorithmically analyzing seed sets. However, addresses lack semantic information and contain numerous addressing schemes, leading difficulty designing effective algorithms. In this paper, we introduce our approach 6VecLM explore achieving such generation The...
Encoding domain knowledge into the prior over high-dimensional weight space of a neural network is challenging but essential in applications with limited data and weak signals. Two types are commonly available scientific applications: 1. feature sparsity (fraction features deemed relevant); 2. signal-to-noise ratio, quantified, for instance, as proportion variance explained (PVE). We show how to encode both widely used Gaussian scale mixture priors Automatic Relevance Determination....
With the continuous growth of video traffic on whole Internet, it is vital importance for content providers to provide high-quality service. Monitoring QoE (Quality Experience) can help them improve their However, most encrypted which has unreadable payload and less information in transport layer that makes monitoring a challenging task. measured from multiple metrics many methods have been proposed infer it. Traditional machine learning based two main shortcomings. 1) These ignore relevance...