A5101559514- Adversarial Robustness in Machine Learning
- Anomaly Detection Techniques and Applications
- Biometric Identification and Security
- Integrated Circuits and Semiconductor Failure Analysis
- Face recognition and analysis
- Domain Adaptation and Few-Shot Learning
- Physical Unclonable Functions (PUFs) and Hardware Security
- Advanced Neural Network Applications
- Advanced Graph Neural Networks
- Bacillus and Francisella bacterial research
- Spam and Phishing Detection
- Gait Recognition and Analysis
- Railway Engineering and Dynamics
- Transport and Logistics Innovations
- Transportation Systems and Logistics
- Text and Document Classification Technologies
- Industrial Vision Systems and Defect Detection
- Brain Tumor Detection and Classification
- Railway Systems and Energy Efficiency
- Autonomous Vehicle Technology and Safety
- Data Visualization and Analytics
- Digital Media Forensic Detection
- Telecommunications and Broadcasting Technologies
- Parallel Computing and Optimization Techniques
- Internet Traffic Analysis and Secure E-voting
China Academy of Railway Sciences
2022-2024
University of Maryland, College Park
2024
Shanghai Jiao Tong University
2019-2022
National University of Singapore
2022
Fudan University
2022
Institute of Automation
2021
Chinese Academy of Sciences
2021
University of Rochester
2019-2020
The University of Texas at Austin
2015-2016
Recently, researchers have started decomposing deep neural network models according to their semantics or functions. Recent work has shown the effectiveness of decomposed functional blocks for defending adversarial attacks, which add small input perturbation image fool DNN models. This proposes a profiling-based method decompose different blocks, lead effective path as new approach exploring DNNs' internal organization. Specifically, per-image can be aggregated class-level path, through we...
Enterprise Web applications are moving towards server-side scripting using managed languages. Within this shifting context, event-driven programming is emerging as a crucial model to achieve scalability. In paper, we study the microarchitectural implications of scripting, JavaScript in particular, from unique perspective. Using Node.js framework, come several critical conclusions. First, unlike traditional server-workloads such CloudSuite and BigDataBench that based on conventional...
Deep learning is vulnerable to adversarial attacks, where carefully-crafted input perturbations could mislead a well-trained Neural Network (DNN) produce incorrect results. Adversarial attacks jeopardize the safety, security, and privacy of DNN-enabled systems. Today's countermeasures either do not have capability detect samples at inference-time, or introduce prohibitively high overhead be practical inference-time.We propose Ptolemy, an algorithm-architecture co-designed system that detects...
Occlusion is a common problem with biometric recognition in the wild. The generalization ability of CNNs greatly decreases due to adverse effects various occlusions. To this end, we propose novel unified framework integrating merits both and graph models overcome occlusion problems recognition, called multiscale dynamic representation (MS-DGR). More specifically, group deep features reflected on certain subregions recrafted into feature (FG). Each node inside FG deemed characterize specific...
With the widespread use of face masks due to COVID-19 pandemic, accurate masked recognition has become more crucial than ever. While several studies have investigated using convolutional neural networks (CNNs), there is a paucity research exploring plain Vision Transformers (ViTs) for this task. Unlike ViT models used in image classification, object detection, and semantic segmentation, model trained by modern losses struggles converge when from scratch. To end, paper initializes parameters...
Iris segmentation and localization in non-cooperative environment is challenging due to illumination variations, long distances, moving subjects limited user cooperation, etc. Traditional methods often suffer from poor performance when confronted with iris images captured these conditions. Recent studies have shown that deep learning could achieve impressive on task. In addition, as defined an annular region between pupil sclera, geometric constraints be imposed help locating the more...
Deep learning-based face recognition models are vulnerable to adversarial attacks. To curb these attacks, most defense methods aim improve the robustness of against perturbations. However, generalization capacities quite limited. In practice, they still unseen learning fairly robust general perturbations, such as Gaussian noises. A straightforward approach is inactivate perturbations so that can be easily handled this paper, a plug-and-play method, named perturbation inactivation (PIN),...
Purpose In recent years, railway systems worldwide have faced challenges such as the modernization of engineering projects, efficient management intelligent digital equipment, rapid growth in passenger and freight transport demands, customized services ubiquitous safety. The transformation toward railways has emerged an effective response to formidable confronting industry, thereby becoming inevitable global trend development. Design/methodology/approach This paper, therefore, conducts a...
Expert-curated guides to the best of CS research.
Deep neural networks have proven to be highly effective in the face recognition task, as they can map raw samples into a discriminative high-dimensional representation space. However, understanding this complex space proves challenging for human observers. In paper, we propose novel approach that interprets deep models via facial attributes. To achieve this, introduce two-stage framework recovers attributes from representations. This allows us quantitatively measure significance of relation...
Abstract: Because of the increasing prevalence and sophistication credit card theft, standard detection measures frequently fail. This study investigates use several machine learning algorithms to improve fraud in bank transactions. The research aims develop enhanced systems by employing an integrated strategy involving data preprocessing, application various classification algorithms, performance evaluation. thoroughly examines such as Random Forest, Logistic Regression, Neural Networks,...
Convolutional neural networks trained on large datasets can generalize various down-streaming tasks, including industrial anomaly detection and localization, which is critical in modern large-scale manufacturing. Whereas previous methods have demonstrated that the feature fusion strategy across multiple layers effective for better performance they lack flexibility intervening manipulating local global information composition process. Through experiments, we demonstrate brute-force used leads...
Recently, researchers have started decomposing deep neural network models according to their semantics or functions. Recent work has shown the effectiveness of decomposed functional blocks for defending adversarial attacks, which add small input perturbation image fool DNN models. This proposes a profiling-based method decompose different blocks, lead effective path as new approach exploring DNNs' internal organization. Specifically, per-image can be aggregated class-level path, through we...
Deep learning is vulnerable to adversarial attacks, where carefully-crafted input perturbations could mislead a well-trained Neural Network produce incorrect results. Today's countermeasures attacks either do not have capability detect samples at inference time, or introduce prohibitively high overhead be practical time. We propose Ptolemy, an algorithm-architecture co-designed system that detects time with low and accuracy.We exploit the synergies between DNN imperative program execution:...
Our third installment of Research for Practice brings readings spanning programming languages, compilers, privacy, and the mobile web.
The transmission bandwidth of a vehicle–ground connection is low when an EMU (electric multiple unit) running in high-speed scenario. To this end, paper focuses on the need to solve problem poor integration information network, and proposes network scheme for EMUs based small base stations. Based existing wi-fi system EMU, order realize coverage 5G signal carriage, paper—through deployment technical characteristics 5G—sinks customized UPF (user plane function) MEC (mobile edge computing)...
Abstract Deep neural network model extraction attack is the process of retraining a surrogate based on outputs target with given set inputs. Such attacks are hard to defend for sake owners’ interest. Recently, some work propose watermarking scheme image processing networks, which able prove intellectual property deep models even after attack. This makes sure that, once (an network) watermarked, we can extract watermark from output model. In this paper, new fight against latest method....
The spam filtering system is used to identify which emails in the received are completely meaningless recipient and perform operations such as interception deletion. Nowadays, with rapid development of Internet, while e-mail provides convenience for people, also comes along it, brings many troubles users. According statistics, 80% world spam, e-spam really annoying. Therefore, how solve problem has important practical significance. Spam using Bayesian theory a statistical technique applied...
Occlusion is a common problem with biometric recognition in the wild. The generalization ability of CNNs greatly decreases due to adverse effects various occlusions. To this end, we propose novel unified framework integrating merits both and graph models overcome occlusion problems recognition, called multiscale dynamic representation (MS-DGR). More specifically, group deep features reflected on certain subregions recrafted into feature (FG). Each node inside FG deemed characterize specific...
As the informatization, digitization, and intelligence of railways continues to progress, new advantages, such as massive data resources rich application scenarios, promote generation autonomous intelligent high-speed railway system (AIHSRS). Based on an analysis current state transportation research, this article proposes overall architecture AIHSRS, with its connotations characteristics. Specifically, fundamental platform AIHSRS consists entity model layer, fusion mechanism interface...