A5072697059- Domain Adaptation and Few-Shot Learning
- Neurobiology and Insect Physiology Research
- Multimodal Machine Learning Applications
- Anomaly Detection Techniques and Applications
- Neural Networks and Reservoir Computing
- Insect and Arachnid Ecology and Behavior
- Neural Networks and Applications
- Cancer-related molecular mechanisms research
- Respiratory viral infections research
- Advanced Fluorescence Microscopy Techniques
- Morphological variations and asymmetry
- Topic Modeling
Zhejiang University
2020-2024
Zhejiang Lab
2024
Zhejiang University of Science and Technology
2023
Domain adaptation methods reduce domain shift typically by learning domain-invariant features. Most existing are built on distribution matching, e.g., adversarial adaptation, which tends to corrupt feature discriminability. In this paper, we propose Discriminative Radial Adaptation (DRDR) bridges source and target domains via a shared radial structure. It's motivated the observation that as model is trained be progressively discriminative, features of different categories expand outwards in...
Channel attention mechanisms endeavor to recalibrate channel weights enhance representation abilities of networks. However, mainstream methods often rely solely on global average pooling as the feature squeezer, which significantly limits overall potential models. In this paper, we investigate statistical moments maps within a neural network. Our findings highlight critical role high-order in enhancing model capacity. Consequently, introduce flexible and comprehensive mechanism termed...
Existing machine learning (ML) models are often fragile in open environments because the data distribution frequently shifts. To address this problem, domain generalization (DG) aims to explore underlying invariant patterns for stable prediction across domains. In work, we first characterize that failure of conventional ML DG attributes an inadequate identification causal structures. We further propose a novel Directed Acyclic Graph (dubbed iDAG) searching framework attains graphical...
Distributional shift between domains poses great challenges to modern machine learning algorithms. The domain generalization (DG) signifies a popular line targeting this issue, where these methods intend uncover universal patterns across disparate distributions. Noted, the crucial challenge behind DG is existence of irrelevant features, and most prior works overlook information. Motivated by this, we propose novel contrastive-based disentanglement method CDDG, effectively utilize...
Channel attention mechanisms endeavor to recalibrate channel weights enhance representation abilities of networks. However, mainstream methods often rely solely on global average pooling as the feature squeezer, which significantly limits overall potential models. In this paper, we investigate statistical moments maps within a neural network. Our findings highlight critical role high-order in enhancing model capacity. Consequently, introduce flexible and comprehensive mechanism termed...
Black-box domain adaptation (BDA) targets to learn a classifier on an unsupervised target while assuming only access black-box predictors trained from unseen source data. Although few BDA approaches have demonstrated promise by manipulating the transferred labels, they largely overlook rich underlying structure in domain. To address this problem, we introduce novel separation and alignment framework for BDA. Firstly, locate those well-adapted samples via loss ranking flexible...
Domain adaptation (DA) aims to transfer discriminative features learned from source domain target domain. Most of DA methods focus on enhancing feature transferability through domain-invariance learning. However, source-learned discriminability itself might be tailored biased and unsafely transferable by spurious correlations, \emph{i.e.}, part source-specific are correlated with category labels. We find that standard learning suffers such correlations incorrectly transfers the...
Abstract How muscle actions are coordinated to realize animal movement is a fundamental question in behavioral study. To obtain the overall muscular activity patterns accompanying behaviors at high spatiotemporal resolution technically difficult. In this work, we used light sheet microscopy simultaneously image and analyze activity, length orientation of Drosophila larval muscles across body segments single nearly free behaviors. For typical modes such as peristalsis, head cast turning,...