A5015529434- Software Testing and Debugging Techniques
- Machine Learning and Algorithms
- Adversarial Robustness in Machine Learning
- Advanced Technologies in Various Fields
- Formal Methods in Verification
- Machine Learning and Data Classification
- Smart Grid Security and Resilience
- Artificial Intelligence in Healthcare
- Healthcare Systems and Public Health
- Advanced Neural Network Applications
- Information and Cyber Security
- Network Security and Intrusion Detection
- Anomaly Detection Techniques and Applications
- Hand Gesture Recognition Systems
- Software Reliability and Analysis Research
- Software Engineering Research
Microsoft Research Asia (China)
2023-2024
NARI Group (China)
2024
Singapore University of Technology and Design
2015-2017
Due to the great advance in machine learning (ML) techniques, numerous ML models are expanding their application domains recent years. To adapt for resource-constrained platforms such as mobile and Internet of Things (IoT) devices, pre-trained often processed enhance efficiency compactness, using optimization techniques pruning quantization. Similar process other complex systems, e.g., program compilers databases, optimizations can contain bugs, leading severe consequences system crashes...
Convex hulls are commonly used to tackle the non-linearity of activation functions in verification neural networks. Computing exact convex hull is a costly task though. In this work, we propose fast and precise approach over-approximating ReLU function (referred as hull), one most functions. Our key insight formulate polytope that ”wraps” hull, by reusing linear pieces lower faces constructing upper adjacent faces. The can be efficiently constructed based on edges vertices faces, given an n...
Aiming at the low efficiency of gesture interaction in command and control system, this paper proposes a optimization method based on complexity. Ten joint points are selected to model gestures, pairs complexity defined. Complexity is used as basis for optimization. Finally, commonly GOMS human factors engineering evaluate efficiency. The results show that can reduce task completion time effectively improve
We introduce a fast invariant inference framework based on active learning and SVMs (Support Vector Machines) which aims to systematically generate variety of loop invariants efficiently. Given program containing one along with precondition post-condition, our approach can learn an is sufficiently strong for verification or otherwise provide counter-examples assist software developers locate bugs. By invoking checking phases iteratively, preliminary experiments show, this may be potentially...
Recently, several abstraction refinement techniques have been proposed to improve the verification precision for deep neural networks (DNNs). However, these usually take many steps verify a property and decision in each step is hard interpret, thus hindering their analysis, reasoning optimization.In this work, we propose SURGEON, novel DNN approach that both effective interpretable, allowing analyst understand why how made. The main insight leverage 'interpretable' nature of debugging...