A5101864441- Adversarial Robustness in Machine Learning
- Software Testing and Debugging Techniques
- Advanced Neural Network Applications
- Digital and Cyber Forensics
- Privacy-Preserving Technologies in Data
- CCD and CMOS Imaging Sensors
- Machine Learning and Algorithms
- Biochemical effects in animals
- Protein Hydrolysis and Bioactive Peptides
- Anomaly Detection Techniques and Applications
- Sparse and Compressive Sensing Techniques
- Computational Drug Discovery Methods
- Photoacoustic and Ultrasonic Imaging
- Risk and Safety Analysis
- Fault Detection and Control Systems
- Software System Performance and Reliability
- Software Reliability and Analysis Research
Zhejiang University
2021-2024
University of Oxford
2024
Nanjing University of Posts and Telecommunications
2022
South China Agricultural University
2019
Recently, there has been a significant growth of interest in applying software engineering techniques for the quality assurance deep learning (DL) systems. One popular direction is testing, where adversarial examples (a.k.a.~bugs) DL systems are found either by fuzzing or guided search with help certain testing metrics. However, recent studies have revealed that commonly used neuron coverage metrics existing approaches not correlated to model robustness. It also an effective measurement on...
Deep learning models, especially those large-scale and high-performance ones, can be very costly to train, demanding a considerable amount of data computational resources. As result, deep models have become one the most valuable assets in modern artificial intelligence. Unauthorized duplication or reproduction lead copyright infringement cause huge economic losses model owners, calling for effective protection techniques. Existing techniques are mostly based on watermarking, which embeds an...
Due to their beneficial effects on human health, antioxidant peptides have attracted much attention from researchers. However, the structure-activity relationships of not been fully understood. In this paper, quantitative (QSAR) models were built two datasets, i.e., ferric thiocyanate (FTC) dataset and ferric-reducing power (FRAP) dataset, containing 214 172 unique tripeptides, respectively. Sixteen amino acid descriptors used model population analysis (MPA) was then applied improve QSAR for...
Recently, there has been significant growth of interest in applying software engineering techniques for the quality assurance deep learning (DL) systems. One popular direction is DL testing—that is, given a property test, defects systems are found either by fuzzing or guided search with help certain testing metrics. However, recent studies have revealed that neuron coverage metrics, which commonly used most existing approaches, not necessarily correlated model (e.g., robustness, studied...
Deep learning (DL) models, especially those large-scale and high-performance ones, can be very costly to train, demanding a great amount of data computational resources. Unauthorized reproduction DL models lead copyright infringement cause huge economic losses model owners. Existing protection techniques are mostly based on watermarking, which embeds an owner-specified watermark into the model. While being able provide exact ownership verification, these 1) invasive, as they need tamper with...
Due to the vast testing space, increasing demand for effective and efficient of deep neural networks (DNNs) has led development various DNN test case prioritization techniques. However, fact that DNNs can deliver high-confidence predictions incorrectly predicted examples, known as over-confidence problem, causes these methods fail reveal errors. To address this limitation, in work, we propose FAST, a method boosts existing through guided FeAture SelecTion. FAST is based on insight certain...
Deep learning (DL) models have become one of the most valuable assets in modern society, and those complex ones require millions dollars for model development. As a result, unauthorized duplication or reproduction DL can lead to copyright infringement cause huge economic losses owners. In this work, we present Deepjudge, testing framework protection. Judgequantitatively tests similarities between two models: victim suspect model. It leverages diverse set metrics efficient test case...
As a remarkable advancement in signal processing, compressed sensing (CS) has proven to be valuable source-limited applications, such as magnetic resonance imaging and computational imaging. Due its capability meet the requirements for real-time deep learning-based reconstruction block-based image CS recently become hot topic. However, existing methods suffer from blocking artifacts, have limited performance due ignoring lost information during measurement. To address these problems, this...
Recently, there has been a significant growth of interest in applying software engineering techniques for the quality assurance deep learning (DL) systems. One popular direction is testing, where adversarial examples (a.k.a.~bugs) DL systems are found either by fuzzing or guided search with help certain testing metrics. However, recent studies have revealed that commonly used neuron coverage metrics existing approaches not correlated to model robustness. It also an effective measurement on...