A5100665185- Domain Adaptation and Few-Shot Learning
- Multimodal Machine Learning Applications
- Quantum Information and Cryptography
- Quantum Computing Algorithms and Architecture
- Advanced Neural Network Applications
- Quantum Mechanics and Applications
- Generative Adversarial Networks and Image Synthesis
- Respiratory viral infections research
- Cancer-related molecular mechanisms research
- Radar Systems and Signal Processing
- Adversarial Robustness in Machine Learning
- Face recognition and analysis
- Molecular Communication and Nanonetworks
- Advanced Image and Video Retrieval Techniques
- Retinal and Optic Conditions
- Cell Image Analysis Techniques
- Retinal Imaging and Analysis
- Digital Imaging for Blood Diseases
- Advanced Mathematical Modeling in Engineering
- Advanced Image Processing Techniques
- Visual Attention and Saliency Detection
- Wireless Signal Modulation Classification
- COVID-19 diagnosis using AI
- Anomaly Detection Techniques and Applications
- Advanced Neuroimaging Techniques and Applications
Nankai University
2022-2024
Ningxia University
2023
Harbin Institute of Technology
2019-2022
Universitat Autònoma de Barcelona
2021
Nanjing University of Science and Technology
2021
Domain adaptation (DA) aims to transfer the knowledge learned from a source domain an unlabeled target domain. Some recent works tackle source-free (SFDA) where only pre-trained model is available for However, those methods do not consider keeping performance which of high practical value in real world applications. In this paper, we propose new paradigm called Generalized Source-free Adaptation (G-SFDA), needs perform well on both and domains, with access current data during adaptation....
Domain adaptation (DA) aims to alleviate the domain shift between source and target domain. Most DA methods require access data, but often that is not possible (e.g. due data privacy or intellectual property). In this paper, we address challenging source-free (SFDA) problem, where pretrained model adapted in absence of data. Our method based on observation which might no longer align with classifier, still forms clear clusters. We capture intrinsic structure by defining local affinity...
We propose a simple but effective source-free domain adaptation (SFDA) method. Treating SFDA as an unsupervised clustering problem and following the intuition that local neighbors in feature space should have more similar predictions than other features, we to optimize objective of prediction consistency. This encourages neighborhood features while farther away dissimilar predictions, leading efficient cluster assignment simultaneously. For training, seek upper-bound resulting two terms....
Background: RETFound, a self-supervised, retina-specific foundation model (FM), showed potential in downstream applications. However, its comparative performance with traditional deep learning (DL) models remains incompletely understood. This study aimed to evaluate RETFound against three ImageNet-pretrained supervised DL (ResNet50, ViT-base, SwinV2) detecting ocular and systemic diseases. Methods: We fine-tuned/trained on full datasets, 50%, 20%, fixed sample sizes (400, 200, 100 images,...
Domain adaptation (DA) aims to alleviate the domain shift between source and target domain. Most DA methods require access data, but often that is not possible (e.g., due data privacy or intellectual property). In this paper, we address challenging source-free (SFDA) problem, where pretrained model adapted in absence of data. Our method based on observation which might align with classifier, still forms clear clusters. We capture intrinsic structure by defining local affinity encourage label...
Quantum key distribution (QKD) can provide point-to-point information-theoretic secure services for two connected users. In fact, the development of QKD networks needs more focus from scientific community in order to broaden service scale technology deliver end-to-end services. Of course, some recent efforts have been made develop communication protocols based on QKD. However, due limited generation capability devices, high quantum utilization is major concern networks. Since traditional...
Due to the intrinsic point-to-point characteristic of quantum key distribution (QKD) systems, it is necessary study and develop QKD network technology provide a secure communication service for large-scale nodes over large area. Considering quality assurance required such cost limitations, building an effective mathematical model becomes critical task. In this paper, flow-based proposed describe using concepts language. addition, investigation on topology evaluation was conducted unique...
Domain generalization in semantic segmentation aims to alleviate the performance degradation on unseen domains through learning domain-invariant features. Existing methods diversify images source domain by adding complex or even abnormal textures reduce sensitivity domain-specific However, these approaches depends heavily richness of texture bank and training them can be time-consuming. In contrast importing arbitrarily augmenting styles randomly, we focus single itself achieve...
With the rapid development of 5G technology, many innovative technologies and various standards have emerged in field wireless communication. Modulation Recognition (MR) is one important communication system, whose main purpose to recognize modulation mode received signal overcome influence noise interference. In order solve problem recognition accuracy under low signal-to-noise ratio, an instantaneous feature fusion algorithm based on deep learning framework proposed. This method uses...
We address the source-free domain adaptation (SFDA) problem, where only source model is available during to target domain. consider two settings: offline setting all data can be visited multiple times (epochs) arrive at a prediction for each sample, and online needs directly classified upon arrival. Inspired by diverse classifier based methods, in this paper we introduce second classifier, but with another head fixed. When adapting domain, additional initialized from expected find...
One of the key components within diffusion models is UNet for noise prediction. While several works have explored basic properties decoder, its encoder largely remains unexplored. In this work, we conduct first comprehensive study encoder. We empirically analyze features and provide insights to important questions regarding their changes at inference process. particular, find that change gently, whereas decoder exhibit substantial variations across different time-steps. This finding inspired...
Segmentation of multiple surfaces in medical images is a challenging problem, further complicated by the frequent presence weak boundary and mutual influence between adjacent objects. The traditional graph-based optimal surface segmentation method has proven its effectiveness with ability capturing various priors uniform graph model. However, efficacy heavily relies on handcrafted features that are used to define cost for "goodness" surface. Recently, deep learning (DL) emerging as powerful...
In this paper, we investigate Source-free Open-partial Domain Adaptation (SF-OPDA), which addresses the situation where there exist both domain and category shifts between source target domains. Under SF-OPDA setting, aims to address data privacy concerns, model cannot access anymore during adaptation. We propose a novel training scheme learn (n+1)-way classifier predict n classes unknown class, samples of only known categories are available for training. Furthermore, adaptation, simply...
We address the source-free domain adaptation (SFDA) problem, where only source model is available during to target domain. consider two settings: offline setting all data can be visited multiple times (epochs) arrive at a prediction for each sample, and online needs directly classified upon arrival. Inspired by diverse classifier based methods, in this paper we introduce second classifier, but with another head fixed. When adapting domain, additional initialized from expected find...
As the Quantum Key Distribution (QKD) technology supporting pointto-point application matures, need to build Secure Communication Network (QSCN) guarantee security of a large scale nodes becomes urgent. Considering project time and expense control, it is first choice QSCN based on an existing classical network. Suitable modeling simulation are very important construct successfully efficiently. In this paper, practical model, which can reflect network state well, proposed. The model considers...
Domain adaptation (DA) aims to transfer the knowledge learned from a source domain an unlabeled target domain. Some recent works tackle source-free (SFDA) where only pre-trained model is available for However, those methods do not consider keeping performance which of high practical value in real world applications. In this paper, we propose new paradigm called Generalized Source-free Adaptation (G-SFDA), needs perform well on both and domains, with access current data during adaptation....
Domain generalization in semantic segmentation aims to alleviate the performance degradation on unseen domains through learning domain-invariant features. Existing methods diversify images source domain by adding complex or even abnormal textures reduce sensitivity specific However, these approaches depend heavily richness of texture bank, and training them can be time-consuming. In contrast importing arbitrarily augmenting styles randomly, we focus single itself achieve generalization. this...
Domain adaptation (DA) aims to alleviate the domain shift between source and target domain. Most DA methods require access data, but often that is not possible (e.g. due data privacy or intellectual property). In this paper, we address challenging source-free (SFDA) problem, where pretrained model adapted in absence of data. Our method based on observation which might align with classifier, still forms clear clusters. We capture intrinsic structure by defining local affinity encourage label...
With the growing complexity of quantum key distribution (QKD) network structures, aforehand topology design is great significance to support a large-number nodes over large-spatial area. However, exclusivity channels, limitation generation capabilities, variety QKD protocols and necessity untrusted-relay selection, make optimal very complicated task. In this research, hybrid studied for first time from perspective topology, by analyzing topological differences various protocols. addition,...