A5100639554- Text and Document Classification Technologies
- Adversarial Robustness in Machine Learning
- Privacy-Preserving Technologies in Data
- Stochastic Gradient Optimization Techniques
- Fault Detection and Control Systems
- Software Engineering Techniques and Practices
- Advanced Graph Neural Networks
- Anomaly Detection Techniques and Applications
- Software Reliability and Analysis Research
- Topic Modeling
- Web Data Mining and Analysis
- Cancer Cells and Metastasis
- Domain Adaptation and Few-Shot Learning
- Random Matrices and Applications
- Hydrogen embrittlement and corrosion behaviors in metals
- Advanced Malware Detection Techniques
- Service-Oriented Architecture and Web Services
- Statistical Methods and Inference
- Diverse Education Studies and Reforms
- Bayesian Modeling and Causal Inference
- Advanced Text Analysis Techniques
- Natural Language Processing Techniques
- Nuclear Engineering Thermal-Hydraulics
- Cytokine Signaling Pathways and Interactions
- Data Quality and Management
China University of Petroleum, East China
2024
Sun Yat-sen University
2023
Xi'an Jiaotong University
2022
Shenyang Aerospace University
2022
Tianjin Medical University Cancer Institute and Hospital
2020
University of Minnesota System
2020
Rutgers Sexual and Reproductive Health and Rights
2019
Beijing University of Posts and Telecommunications
2015
Université de Toulouse
2010-2012
Roche (France)
2012
To achieve reliable and automatic anomaly detection (AD) for large equipment such as liquid rocket engine (LRE), multisource data are commonly manipulated in deep learning pipelines. However, current AD methods mainly aim at single source or modality, whereas existing multimodal cannot effectively cope with a common issue, modality incompleteness. this end, we propose an unsupervised method missing sources LRE system. The proposed handles intramodality fusion, intermodality decision fusion...
With the rapid development of Web2.0, more and people like to show their life or opinions on social media websites forums, such as Weibo, Twitter Tianya, which produce masses short texts. In order manage these texts effectively, Short Text Classification becomes an important branch Classification. However, because text length, lack signals, sparseness features, it is very difficult achieve high quality classification by using conventional methods. This paper proposes a novelty feature...
Although a number of studies are devoted to novel category discovery, most them assume static setting where both labeled and unlabeled data given at once for finding new categories. In this work, we focus on the application scenarios continuously fed into discovery system. We refer it as {\bf Continuous Category Discovery} ({\bf CCD}) problem, which is significantly more challenging than setting. A common challenge faced by that different sets features needed classification discovery: class...
Entity Set Expansion (ESE) is a valuable task that aims to find entities of the target semantic class described by given seed entities. Various Natural Language Processing (NLP) and Information Retrieval (IR) downstream applications have benefited from ESE due its ability discover knowledge. Although existing corpus-based methods achieved great progress, they still rely on corpora with high-quality entity information annotated, because most them need obtain context patterns through position...
Federated learning (FL) is a recently proposed distributed machine paradigm dealing with and private data sets. Based on the partition pattern, FL often categorized into horizontal, vertical, hybrid settings. Despite fact that many works have been developed for first two approaches, setting (which deals partially overlapped feature space sample space) remains less explored, though this extremely important in practice. In paper, we set up new model-matching-based problem formulation FL, then...
Entity Set Expansion (ESE) is a valuable task that aims to find entities of the target semantic class described by given seed entities. Various Natural Language Processing (NLP) and Information Retrieval (IR) downstream applications have benefited from ESE due its ability discover knowledge. Although existing corpus-based methods achieved great progress, they still rely on corpora with high-quality entity information annotated, because most them need obtain context patterns through position...
Process-control systems are usually real time and embedded within the larger environment. Controllers required to ensure desired environment properties through performing ongoing control behavior at interface with sensors actuators. However, is heterogeneous informal, which poses special challenge develop complete safe specification. Problem Frames a rigor requirements analysis approach for software problems emphasize fundamental role of in capturing In this paper, we contribute exploring...
Purpose The purpose of this study is to characterize the galvanic corrosion behavior a simulated X80 pipeline steel welded joint (PSWJ) reconstructed by wire beam electrode (WBE) and numerical simulation methods. Design/methodology/approach an PSWJ was studied using WBE microstructures coarse-grained heat affected zone, fine-grained zone intercritical were in via Gleeble thermomechanical processing. Findings Comparing current density coupled isolated weld metal (WM), base (BM) heat-affected...
Vision-language models (VLMs) have significantly advanced autonomous driving (AD) by enhancing reasoning capabilities. However, these remain highly vulnerable to adversarial attacks. While existing research has primarily focused on general VLM attacks, the development of attacks tailored safety-critical AD context been largely overlooked. In this paper, we take first step toward designing specifically targeting VLMs in AD, exposing substantial risks pose within critical domain. We identify...
Meaning of expectations and differences among needs, requirements are ambiguous in literatures practice. In this paper, we contribute to give a possible clarification ambiguity. The relationships examined together with characterization customer expectations. We also introduce the elicitation process based on engineering approach, which provides an explicit controllable corresponding quality model. Our goal is finally connect value dimensions enable engineering.
Abstract Proper management of requirement traceability information is vital for the project success. Often, it debatable what to trace and how, without burdening engineer with useless information. Requirement engineering in context hardware intensive systems employs different techniques recovery, management. In this paper, we have tried address problem systems' engineering. We classified information, which needs be traced eight categories, distinguished their usage various activities...
We develop new approaches in multi-class settings for constructing proper scoring rules and hinge-like losses establishing corresponding regret bounds with respect to the zero-one or cost-weighted classification loss. Our construction of involves deriving inverse mappings from a concave generalized entropy loss through use convex dissimilarity function related multi-distribution $f$-divergence. Moreover, we identify classes rules, which also recover reveal interesting relationships between...
Differentially Private Stochastic Gradient Descent with gradient clipping (DPSGD-GC) is a powerful tool for training deep learning models using sensitive data, providing both solid theoretical privacy guarantee and high efficiency. However, DPSGD-GC to ensure Differential Privacy (DP) comes at the cost of model performance degradation due DP noise injection clipping. Existing research has extensively analyzed convergence DPSGD-GC, shown that it only converges when large thresholds are...
Pre-trained vision models (PVMs) have become a dominant component due to their exceptional performance when fine-tuned for downstream tasks. However, the presence of backdoors within PVMs poses significant threats. Unfortunately, existing studies primarily focus on backdooring classification task, neglecting potential inherited in tasks such as detection and segmentation. In this paper, we propose Trojan attack, which embeds into PVM, enabling attacks across various We highlight challenges...
Consider semi-supervised learning for classification, where both labeled and unlabeled data are available training. The goal is to exploit datasets achieve higher prediction accuracy than just using alone. We develop a logistic method based on exponential tilt mixture models, by extending statistical equivalence between regression modeling. study maximum nonparametric likelihood estimation derive novel objective functions which shown be Fisher consistent. also propose regularized construct...
In view of the high dimensionality data in many fields and serious multiple correlation between variables, this paper proposes an interpretable partial least square regression (PLSR) modeling method. Compared with principal component (PCR), when there are a large number predictors, both PLSR PCR model response predictors highly correlated or even collinear. Both these methods construct new (called components) as linear combinations original but they components different ways. We use series...