A5100652479- Advanced Neural Network Applications
- Topic Modeling
- Industrial Vision Systems and Defect Detection
- Generative Adversarial Networks and Image Synthesis
- Face recognition and analysis
- Image and Object Detection Techniques
- Mineral Processing and Grinding
- Adversarial Robustness in Machine Learning
- Multimodal Machine Learning Applications
- Natural Language Processing Techniques
- Advanced Vision and Imaging
- Advanced X-ray and CT Imaging
- Geochemistry and Geologic Mapping
- Advanced Image Processing Techniques
- Face and Expression Recognition
- Human Pose and Action Recognition
- Medical Image Segmentation Techniques
- Machine Learning in Bioinformatics
- Advanced Malware Detection Techniques
- Domain Adaptation and Few-Shot Learning
- Remote-Sensing Image Classification
- Advanced Text Analysis Techniques
- Advanced Computational Techniques and Applications
- Video Surveillance and Tracking Methods
- Advanced Image and Video Retrieval Techniques
Jiangnan University
2025
Shanghai University of Traditional Chinese Medicine
2023-2024
China University of Geosciences
2022-2024
Beihang University
2016-2023
Academy of Broadcasting Science
2022-2023
Norwegian University of Science and Technology
2022-2023
Shanghai University
2023
Blood Center of Zhejiang Province
2023
Communication University of China
2017-2022
Guangxi University of Finance and Economics
2022
A fast-paced development of DeepFake generation techniques challenge the detection schemes designed for known type DeepFakes. reliable Deepfake approach must be agnostic to types, which can present diverse quality and appearance. Limited generalizability across different will restrict wide-scale deployment detectors if they fail handle unseen attacks in an open set scenario. We propose a new approach, Multi-Channel Xception Attention Pairwise Interaction (MCX-API), that exploits power...
DeepFakes detection approaches have to be agnostic across generation type, quality, and appearance provide a generalizable detector. Limited generalizability will hinder wide-scale deployment of detectors if they cannot handle unseen attacks in an open set scenario. We propose model that can detect novel unknown/unseen using supervised contrastive (SupCon) loss. As resemble the original image/video greater extent terms it becomes challenging secern them, we exploit contrasts representation...
Abstract Background Automatic segmentation of temporal bone structures from patients' conventional computed tomography (CT) data plays an important role in the image‐guided cochlear implant surgery. Existing convolutional neural network approaches have difficulties segmenting such small tubular structures. Methods We propose a light‐weight three‐dimensional referred to as W‐Net achieve multiobjective including labyrinth, ossicular chain and facial nerve CT images. Data augmentation with...
ABSTRACT Background Attention deficit hyperactivity disorder (ADHD) is one of the most prevalent neurodevelopmental disorders, with an estimated prevalence 5.6% among Chinese children. This population needs to seek novel non‐pharmacological alternatives for treatment. Objective research explored effects 6‐week high‐frequency repetitive transcranial magnetic stimulation (rTMS) on clinical symptoms and executive function in children from 6 10 years ADHD. Methods Forty‐eight diagnosed ADHD were...
Abstract The current method for inspecting microholes in printed circuit boards (PCBs) involves preparing slices followed by optical microscope measurements. However, this approach suffers from low detection efficiency, poor reliability, and insufficient measurement stability. Micro-CT enables the observation of internal structures sample without need slicing, thereby presenting a promising new assessing quality PCBs. This study integrates computer vision technology with computed tomography...
To improve the accessibility of smart devices and to simplify their usage, building models which understand user interfaces (UIs) assist users complete tasks is critical. However, unique challenges are proposed by UI-specific characteristics, such as how effectively leverage multimodal UI features that involve image, text, structural metadata achieve good performance when high-quality labeled data unavailable. address we introduce UIBert, a transformer-based joint image-text model trained...
Cloud gaming (CG) has gradually gained popularity. By leveling shared computing resources on the cloud, CG technology allows those without expensive hardware to enjoy AAA games using a low-end device. However, bandwidth requirement for streaming game video is high, which can cause backbone network congestion large-scale deployment and bills. To address this challenge, authors proposed an innovative edge-assisted architecture that collaboratively uses AI-powered foveated rendering (FR)...
An adversarial example is an input transformed by small perturbations that machine learning models consistently misclassify. While there are a number of methods proposed to generate examples for text data, it not trivial assess the quality these examples, as minor (such changing word in sentence) can lead significant shift their meaning, readability and classification label. In this paper, we propose evaluation framework consisting set automatic metrics human guidelines, rigorously based on...
Many methods based on deep learning have achieved amazing results in image sentiment analysis. However, these existing usually pursue high accuracy, ignoring the effect model training efficiency. Considering that when faced with large-scale analysis tasks, accuracy rate often requires long experimental time. In view of weakness, a method can greatly improve efficiency only small fluctuations is proposed, and singular value decomposition (SVD) used to find sparse feature image, which are...
Physiological signals, such as electrocardiogram (ECG) and wrist pulse signals (WPS), play an important role in diagnosing preventing cardiovascular other physiological diseases. Therefore, accurate classification of has become the key to assist physicians diagnosis. However, this field still faces several prominent challenges, including limited availability data, imbalanced datasets, convergence issues with loss functions, need for model architectures capable accurately detecting waveform...
Pixel differences between classes with low contrast in medical image semantic segmentation tasks often lead to confusion category classification, posing a typical challenge for recognition of small targets. To address this challenge, we propose Contrastive Adaptive Augmented Semantic Segmentation Network differentiable pooling function. Firstly, an Contrast Augmentation module is constructed automatically extract local high-frequency information, thereby enhancing details and accentuating...
Traditional pelvis fracture reduction suffers from some disadvantages. Robot-assisted offers promise in solving these problems. However, the reference to guide robot motion is a key issue that must be resolved. In this paper, we propose physical symmetry and virtual plane-based adopt method of registration calculate plane for reference, which were verified via experiments. The results position experiments original similar; both showed errors less than 4[Formula: see text]mm 2.5[Formula:...
Sentiment analysis is a comprehensive research in Natural Language Processing (NLP) which used to target the emotional context of text. In previous experiments, shallow vocabulary and grammatical features were often considered, implicit semantic neglected. order solve this problem, paper combines training ways word vectors (Word2vec Doc2veC TF-IDF model) machine classification model (SVM, I<; NN, etc.) for then extracting features. The main goal use performance Word2vec Doc2vec models...
We propose a new open question answering framework for over knowledge base (KB).Our system uses both curated KB, Freebase, and one that is extracted automatically by an information extraction model, IE KB.Our consists of only layer paraphrase, compared to the three layers used in previous (Fader et al., 2014).However, because more accurately relation triples combined with linked entities from KB our achieves 7% absolute gain F 1 score system.
With the advent of information and 5G era, contemporary social media platforms increasingly support users to send text, voice, image, video other multimedia data at same time, in which images can convey sender's emotional state very simply directly. The classification emotions expressed has become focus research attracted more attention. Convolutional neural network is often used image analysis processing. In this paper, four different models are constructed based on CNN realize calculation...