A5100353779- Software Testing and Debugging Techniques
- Security and Verification in Computing
- Real-Time Systems Scheduling
- Advanced Malware Detection Techniques
- Surface Roughness and Optical Measurements
- Explainable Artificial Intelligence (XAI)
- Software Engineering Research
- Laser Material Processing Techniques
- Advanced Database Systems and Queries
- Machine Learning and Data Classification
- Logic, programming, and type systems
- Data Management and Algorithms
- Library Science and Information Systems
- Cloud Computing and Resource Management
- Machine Learning and Algorithms
- Religion and Sociopolitical Dynamics in Nigeria
- Scientific Computing and Data Management
- Welding Techniques and Residual Stresses
- Archaeological Research and Protection
- Advanced Graph Neural Networks
- Cultural and Historical Studies
- Advanced Data Storage Technologies
- Recommender Systems and Techniques
- Formal Methods in Verification
- Polynomial and algebraic computation
China Institute of Atomic Energy
2025
East China Normal University
2017-2022
Shanghai Key Laboratory of Trustworthy Computing
2019-2022
Currently, the quartz glass–TC4 dissimilar joint has been applied in fields such as radiation environment testing, reactor engineering, and other areas. However, high brittleness of glass thermal mismatch during welding process limit require further development. Thus, a femtosecond laser was employed to perform direct joining these materials under non-optical contact conditions, with aid well-designed clamp optimized process, effect pulse energy on microstructure mechanical properties...
In 1987, Greene introduced the notion of finite field analogue hypergeometric series. this paper, we give a Appell series and obtain some transformation reduction formulas. We also establish generating functions for over fields.
It is a challenging problem for humans to understand the predictions made by sophisticated machine learning techniques. This paper proposes method interpret complex models on individual samples. We propose first transfer generalization abilities from target model be interpreted, such as deep neural networks, GBDT (Gradient Boosting Decision Tree) through knowledge distillation. then present local explanation faithfully ranking relative importance of interpretable predicates explaining an...
In automobile industry, it is a common approach to develop real-time operating systems under some standards. For instance, OSEK/VDX world-wide adopted open standard. Traditional workflow first understand the standard, design and system, then test its conformance finally deploy. There are several issues with traditional workflow, e.g., ambiguities in standards may lead incorrect implementation of real-world systems; difficult check; bug fixing after costly. To remedy situation, this paper, we...
Graph convolutional networks (GCNs) have been successfully applied to learning tasks on graph-structured data. However, most traditional GCNs based graph convolutions assume homophily in graphs, which leads a poor performance when dealing with heterophilic graphs. Although many novel methods recently proposed deal heterophily, the effect of and heterophily classifying node pairs is not clearly separated existing approaches inevitably influences each other. To various types graphs more...
Synthesizing intended programs from user-specified input-output examples, also known as Programming by Examples (PBE), is a challenging problem in program synthesis, and has been applied to wide range of domains. A key challenge PBE efficiently discover user-intended the search space that can be exponentially large. In this work, we propose method for automatic synthesis functional on list manipulation, using offline-trained Recurrent Neural Network (RNN) models guide search. We adopt...
Document layout analysis (DLA) aims to divide a document image into different types of regions. DLA plays an important role in the content understanding and information extraction systems. Exploring method that can use less data for effective training contributes development DLA. We consider Human-in-the-loop (HITL) collaborative intelligence Our approach was inspired by fact HITL push model learn from unknown problems adding small amount based on knowledge. The select key samples using...
Formal verification has been recognized as an essential part in improving the correctness and soundness of OS kernels. The kernels faces many challenges. For example, formal proof always comes with difficulties non-trivial cost. Researchers often have to verify kernel high level programming languages ignore low languages. In this paper, we propose a framework for verifying whole burden. We claim it end-to-end automated purposes of: (1) that consists assembler C by formally modeling programs...
Explainable artificial intelligence becomes vital for human users to understand and trust the decision-making process results of machine learning methods. Unfortunately, most models are black-box algorithms running behind opaque. In this work, we propose an approach interpreting GBDT (Gradient Boosting Decision Tree Explanation) by extracting probabilistic finite-state automata from trained model. Our method is inspired built upon a previous work that extracts RNN (Recurrent Neural...