Is this you? Claim this profile to correct it, add a bio and choose the work people see first.
Claim this profileWorks182 from public data
- A Comprehensive Survey on Test-Time Adaptation Under Distribution Shifts653
This survey categorizes Test-time adaptation into several distinct groups based on the form of test data, namely, test-time domain adaptation, test-time batch adaptation, and online test-time adaptation.
- NÜWA: Visual Synthesis Pre-training for Neural visUal World creAtion373
Compared to several strong baselines, N\"UWA achieves state-of-the-art results on text-to-image generation, text- to-video generation, video prediction, etc, and it also shows surprisingly good zero-shot capabilities onText-guided image and video manipulation tasks.
- Source Data-absent Unsupervised Domain Adaptation through Hypothesis Transfer and Labeling Transfer370
A novel approach called Source HypOthesis Transfer (SHOT), which learns the feature extraction module for the target domain by fitting the target data features to the frozen source classification module (representing classification hypothesis), and which achieves results surpassing or comparable to the state-of-the-arts.
- Deep Spatial Feature Reconstruction for Partial Person Re-identification: Alignment-free Approach341
The proposed method leverages Fully Convolutional Network (FCN) to generate fix-sized spatial feature maps such that pixel-level features are consistent and can decrease the similarity of coupled images from different persons and increase that from the same person.
- Causality Inspired Representation Learning for Domain Generalization301
A general structural causal model is introduced and a Causality Inspired Representation Learning (CIRL) algorithm is proposed that enforces the representations to satisfy the above properties and then uses them to simulate the causal factors, which yields improved generalization ability.
- Domain Adaptation with Auxiliary Target Domain-Oriented Classifier190
A new pseudo-labeling framework called Auxiliary Target Domain-Oriented Classifier (ATDOC) is proposed, which alleviates the classifier bias by introducing an auxiliary classifier for target data only, to improve the quality of pseudo labels.
- 176
- TALL: Thumbnail Layout for Deepfake Video Detection153
A simple yet effective strategy named Thumbnail Layout (TALL), which transforms a video clip into a pre-defined layout to realize the preservation of spatial and temporal dependencies, which is incorporated into Swin Transformer, forming an efficient and effective method TALL-Swin.
- 143
- 136
- 134
- Bi-Classifier Determinacy Maximization for Unsupervised Domain Adaptation133
The proposed Bi-Classifier Determinacy Maximization (BCDM) is designed with a novel classifier determinacy disparity (CDD) metric, which formulates classifier discrepancy as the class relevance of distinct target predictions and implicitly introduces constraint on the target feature discriminability.
- A Balanced and Uncertainty-Aware Approach for Partial Domain Adaptation126
A novel domain adaptation method BA$3$US is proposed that emphasizes uncertain samples and exploits an adaptive weighted complement entropy objective to encourage incorrect classes to have uniform and low prediction scores.
- 124
- DINE: Domain Adaptation from Single and Multiple Black-box Predictors120
A new two-step knowledge adaptation framework called DIstill and fine-tuNE (DINE) is proposed, which distills the knowledge from the source predictor to a customized target model, then fine-tunes the distilled model to further fit the target domain.
- Exploring uncertainty in pseudo-label guided unsupervised domain adaptation120
This paper proposes to progressively increase the number of target training samples and incorporate the uncertainties to accurately characterize both cross-domain distribution discrepancy and other intra-domain relations, and significantly outperforms recent state-of-the-art approaches.
- Semantic Concentration for Domain Adaptation114
Semantic Concentration for Domain Adaptation is proposed, which encourages the model to concentrate on the most principal features via the pair-wise adversarial alignment of prediction distributions, and can be easily integrated into various DA methods as a regularizer to further boost their performance.
- NUWA-Infinity: Autoregressive over Autoregressive Generation for Infinite Visual Synthesis105
NUWA-Infinity is presented, a generative model for infinite visual synthesis, which is defined as the task of generating arbitrarily-sized high-resolution images or long-duration videos and has superior visual synthesis capabilities in terms of resolution and variable-size generation.
- Learning Feature Recovery Transformer for Occluded Person Re-Identification101
A new approach called Feature Recovery Transformer (FRT) to address the two challenges simultaneously, which mainly consists of visibility graph matching and feature recovery transformer, which significantly outperforms state-of-the-art results.
- Why Attentions May Not Be Interpretable?95
It is demonstrated that one root cause of this phenomenon is the combinatorial shortcuts, which means that, in addition to the highlighted parts, the attention weights themselves may carry extra information that could be utilized by downstream models after attention layers.
- 95
- 91
- Towards Understanding and Mitigating Dimensional Collapse in Heterogeneous Federated Learning89
This work proposes {\sc FedDecorr], a novel method that can effectively mitigate dimensional collapse in federated learning that applies a regularization term during local training that encourages different dimensions of representations to be uncorrelated.
- Mimicking the Oracle: An Initial Phase Decorrelation Approach for Class Incremental Learning89
Class-wise Decorrelation (CwD) is proposed that regularizes representations of each class to scatter more uniformly, thus mimicking the model jointly trained with all classes (i.e., the oracle model), and is simple to implement and easy to plug into existing methods.
- Improving Zero-Shot Generalization for CLIP with Synthesized Prompts85
A plug-and-play generative approach to improve existing fine-tuning methods in pretrained vision-language models by introducing a generator that reconstructs the visual features by inputting the synthesized prompts and the corresponding class names to the textual encoder of CLIP.
- Free Lunch for Domain Adversarial Training: Environment Label Smoothing72
It is demonstrated, both experimentally and theoretically, that ELS can improve training stability, local convergence, and robustness to noisy environment labels, and by incorporating ELS with DAT methods, it is able to yield state-of-art results on a wide range of domain generalization/adaptation tasks.
- 63
- A Hard-to-Beat Baseline for Training-free CLIP-based Adaptation62
This paper revisits a classical algorithm, Gaussian Discriminant Analysis (GDA), and applies it to the downstream classification of CLIP, and extends it to base-to-new generalization and unsupervised learning, demonstrating its superiority over competing approaches.
- Mind the Label Shift of Augmentation-based Graph OOD Generalization62
LiSA, which generates label-invariant augmentations to facilitate graph OOD generalization, and introduces a tractable energy-based regularization to enlarge pair-wise distances between the distributions of environments.
- 56
- Out-of-Distribution Detection: A Task-Oriented Survey of Recent Advances54
This survey uniquely review recent advances in OOD detection from the task-oriented perspective for the first time and classify the methods as training-driven or training-agnostic, according to the user’s access to the model.
- Mimic Embedding via Adaptive Aggregation: Learning Generalizable Person Re-identification54
This paper presents a new approach called Mimic Embedding via adapTive Aggregation (META), which surpasses state-of-the-art DG person ReID methods by a large margin and is expected to mimic embedding for a truly unseen target domain.
- Learning Spatiotemporal Inconsistency via Thumbnail Layout for Face Deepfake Detection52
An elegantly simple yet effective strategy named Thumbnail Layout (TALL), which transforms a video clip into a pre-defined layout to realize the preservation of spatial and temporal dependencies, and introduces a graph reasoning block and semantic consistency loss to strengthen TALL.
- Relation-Guided Representation Learning51
This work proposes a new representation learning method that explicitly models and leverages sample relations, which in turn is used as supervision to guide the representation learning.
- Diagnostic Classification for Human Autism and Obsessive-Compulsive Disorder Based on Machine Learning From a Primate Genetic Model49
The identified core regions may serve as a basis for building markers for ASD and OCD diagnoses, as well as measures of symptom severity, and may inform future development of machine-learning models for psychiatric disorders.
- 48
- 45
- 45
- Adversarial Domain Adaptation With Prototype-Based Normalized Output Conditioner44
This work improves concatenation conditioning by normalizing the output predictions to have the same norm of features, and term the derived method as Normalized OutpUt coNditioner (NOUN), and proposes to condition the cross-domain feature alignment in the prototype space rather than in the output space.
- 43
- Rumor Detection with Diverse Counterfactual Evidence40
This paper addresses the above issues with the proposed Diverse Counterfactual Evidence framework for Rumor Detection (DCE-RD), and designs a subgraph generation strategy to efficiently generate different subgraphs of the event graph that naturally serve as counterfactual evidence for rumor detection.
- 39
- 38
- DiVAE: Photorealistic Images Synthesis with Denoising Diffusion Decoder36
This work proposes a VQ-VAE architecture model with a diffusion decoder (DiVAE) to work as the reconstructing component in image synthesis and explores how to input image embedding into diffusion model for excellent performance and finds that simple modification on diffusion's UNet can achieve it.
- Self-Paced Cross-Modal Subspace Matching36
Experimental results on four multimodal databases show that SCSM outperforms state-of-the-art cross-modal subspace learning methods.
- 35
- Collaboration Equilibrium in Federated Learning35
This paper proposes the concept of benefit graph which describes how each client can benefit from collaborating with other clients and advances a Pareto optimization approach to identify the optimal collaborators and theoretically proves that it can reach a CE from the benefit graph through an iterative graph operation.
- Two Sides of the Same Coin35
This work conducts extensive empirical evaluations to show that fine-tuning effectively enhances model robustness under white-box FGSM attacks and proposes a black-box attack method for transfer learning models which attacks the target model with the adversarial examples produced by its source model.
- 34
- METTL3-mediated m6A modification of HMGA2 mRNA promotes subretinal fibrosis and epithelial–mesenchymal transition34
An epigenetic mechanism of METTL3–m6A–HMGA2 in subretinal fibrosis and EMT of RPE cells, providing a novel therapeutic target for subretina fibrosis secondary to nAMD is identified.
- Domain-Specific Risk Minimization for Domain Generalization34
This paper proposes Domain-specific Risk Minimization (DRM), a generalization bound that explicitly considers the adaptivity gap, and proposes effective gap estimation methods for guiding the selection of a better hypothesis for the target.
- Long-term prediction for high-resolution lane-changing data using temporal convolution network33
A temporal convolutional network is introduced to predict the long-term lane-changing trajectory and behaviour and the advantages of the TCN are rapid response and accurate long- term prediction, which are important for lane- changing assistance in the advanced driver assistance system.
- Dynamic Hypergraph Learning for Collaborative Filtering32
This work proposes a novel dynamic hypergraph learning framework for collaborative filtering (DHLCF), which learns hypergraph structures and makes recommendations collectively in a unified framework and introduces a novel hyper graph learning objective, which forces the learned hypergraphs to retain the original graph topology.
- Heterogeneous Face Recognition via Face Synthesis With Identity-Attribute Disentanglement31
A new HFR method from the perspective of heterogeneous data augmentation, named Face Synthesis with Identity-Attribute Disentanglement (FSIAD) is proposed, which obtains superior performance than previous HFR approaches.
- 30
- 29
- 29
- Benchmarking Test-Time Adaptation against Distribution Shifts in Image Classification27
This work presents a benchmark that systematically evaluates 13 prominent TTA methods and their variants on five widely used image classification datasets and develops a unified framework in PyTorch, which allows for consistent evaluation and comparison of the Tta methods across the different datasets and network architectures.
- Selection Bias Explorations and Debias Methods for Natural Language Sentence Matching Datasets27
This paper investigates the problem of selection bias on six NLSM datasets and finds that four out of them are significantly biased, and proposes a training and evaluation framework to alleviate the bias.
- Additive Adversarial Learning for Unbiased Authentication26
A novel two-stage method that disentangles the class/identity from domain-differences, and it is discussed the necessity to avoid a learning dilemma due to disentangling causally related types ofdomain-difference.
- Adaptive systems on a chip (aSoC) for low-power signal processing26
The use of anAdaptive system-on-a-chip to allow for dynamic power management of video signal processing systems can be exploited to trade quality for low power gracefully.
- A Curriculum-Style Self-Training Approach for Source-Free Semantic Segmentation25
A curriculum-style entropy minimization method is introduced to explore the implicit knowledge from the source model, which fits the trained source model to the target data using certain information from easy-to-hard predictions, which could act as a standard post-processing method for the domain adaptation field.
- Exploring antithrombotic mechanisms and effective constituents of Lagopsis supina using an integrated strategy based on network pharmacology, molecular docking, metabolomics, and experimental verification in rats24
An integrated strategy based on network pharmacology, molecular docking, metabolomics, and in vivo experiments revealed for the first time that L. supina exerts a significant antithrombotic effect through the inhibition of inflammation and coagulation cascade, promotion of angiogenesis, and suppression of platelet activation.
- Adversarial Infidelity Learning for Model Interpretation23
A Model-agnostic Effective Efficient Direct IFS framework for model interpretation, mitigating concerns about sanity, combinatorial shortcuts, model identifiability, and information transmission is proposed and an Adversarial Infidelity Learning (AIL) mechanism is proposed to boost the explanation learning by screening relatively unimportant features.
- 20
- 19
- Research on an Image Denoising Algorithm based on Deep Network Learning19
The structure of noise reduction convolution neural network is further improved, and an image denoising algorithm based on deconvolution neuralNetwork is proposed, and a new loss function calculation method is proposed to optimize the residual learning method.
- 19
- 19
- AesopAgent: Agent-driven Evolutionary System on Story-to-Video Production18
This work proposes AesopAgent, an Agent-driven Evolutionary System on Story-to-Video Production, a practical application of agent technology for multimodal content generation that achieves state-of-the-art performance compared with many previous works in visual storytelling.
- 17
- Reciprocal normalization for domain adaptation17
Reciprocal Normalization is proposed, a novel normalization method that outperforms existing normalization counterparts by a large margin and helps state-of-the-art adaptation approaches achieve better results.
- 17
- MAPS: A Noise-Robust Progressive Learning Approach for Source-Free Domain Adaptive Keypoint Detection17
A unified approach, Mixup Augmentation and Progressive Selection (MAPS), is proposed, to fully exploit the noisy pseudo labels of unlabeled target data during training to fully exploit the noisy pseudo labels of unlabeled target data during training.
- 17
- 16
- Color-Unrelated Head-Shoulder Networks for Fine-Grained Person Re-identification16
A color-unrelated head-shoulder network (CUHS) is developed, and the results show that CUHS surpasses previous methods on both the FG and conventional person re-id problems.
- Informative Data Mining for One-shot Cross-Domain Semantic Segmentation15
This work proposes a novel framework called Informative Data Mining (IDM) that enables efficient one-shot domain adaptation for semantic segmentation and achieves a new state-of-the-art one-shot performance of 56.7%/55.4% on the GTA5/SYNTHIA to Cityscapes adaptation tasks, respectively.
- Finding Diverse and Predictable Subgraphs for Graph Domain Generalization15
This paper proposes a new graph domain generalization framework, dubbed as DPS, by constructing multiple populations from the source domains, which is expected to generalize well to unseen target domains.
- 14
- PANDA: Prototypical Unsupervised Domain Adaptation.14
This work attempts to calibrate the noisy pseudo labels with prototypes, and develops an unified framework termed PrototypicAl uNsupervised Domain Adaptation (PANDA) for UDA, which achieves state-of-the-art or competitive results on multiple UDA benchmarks including both object recognition and semantic segmentation tasks.
- Adapting Vision-Language Models Without Labels: A Comprehensive Survey12
A taxonomy based on the availability and nature of unlabeled visual data is proposed, categorizing existing approaches into four key paradigms: Data-Free Transfer, Unsupervised Domain Transfer, Episodic Test-Time Adaptation, Episodic Test-Time Adaptation, and Online Test-Time Adaptation.
- 12
- 12
- Frustratingly Easy Cross-Modal Hashing12
A simple yet effective cross-modal hashing approach that can be implemented in just three lines of code that is non-iterative and parameter-free, which makes it more attractive for many real-world applications.
- Do We Really Need Curated Malicious Data for Safety Alignment in Multi-modal Large Language Models?11
Experiments show that, without the need for labor-intensive collection of high-quality malicious data, model safety can still be significantly improved, as long as a specific fraction of rejection data exists in the finetuning set, indicating the security alignment is not lost but rather obscured during multi-modal pretraining or instruction finetuning.
- Adversarial Filtering Modeling on Long-term User Behavior Sequences for Click-Through Rate Prediction11
A novel Adversarial Filtering Model (ADFM) is proposed to model long-term user behavior that uses a hierarchical aggregation representation to compress raw behavior sequence and then learns to remove useless behavior information with an adversarial filtering mechanism.
- TAT: Task-Adaptive Transformer for All-in-One Medical Image Restoration10
A task-adaptive Transformer (TAT) is proposed, a novel framework that dynamically adapts to different tasks through two key innovations that mitigate task interference by generating task-specific weight parameters for each task, thereby eliminating potential gradient conflicts on shared weight parameters.
- Improving Generalization with Domain Convex Game10
This work proposes a new perspective on DG that recasts it as a convex game between domains, and first encourages each diversified domain to enhance model generalization by elaborately designing a regularization term based on supermodularity.
- 10
- 10
- Coastal wetland hyperspectral classification under the collaborative of subspace partition and infinite probabilistic latent graph ranking9
The study developed a hyperspectral classification algorithm for coastal wetlands using a combination of subspace partitioning and infinite probabilistic latent graph ranking in a random patch network (the SSP-IPLGR-RPnet model), which provided a better classification performance and was more robust with limited training data compared with that of several state-of-the-art methods.
- 8
- 8
- 8
- AdaptGuard: Defending Against Universal Attacks for Model Adaptation7
This paper studies the vulnerability to universal attacks transferred from the source domain during model adaptation algorithms due to the existence of malicious providers, and proposes a model preprocessing framework, named AdaptGuard, to improve the security of model adaptationgorithms.
- 7
- 7
- 6
- 6
- 6
- 6
- X-GACMN: An X-Shaped Generative Adversarial Cross-Modal Network with Hypersphere Embedding6
This paper introduces a novel X-Shaped Generative Adversarial Cross-Modal Network (X-GACMN), which combines the process of synthetic data generation and distribution adapting into a unified framework to make sure the heterogeneous modality distributions similar to each other in the learned common subspace.
- 5
- 4
- Test-Time Immunization: A Universal Defense Framework Against Jailbreaks for (Multimodal) Large Language Models4
This work proposes a universal defense framework, termed Test-time IMmunization (TIM), which can adaptively defend against various jailbreak attacks in a self-evolving way and implements safety fine-tuning using the detected jailbreak instructions paired with refusal answers.
- Research on intelligent control and emergency response system for power cable tunnels based on AI image recognition and ultra-wideband technology4
The system aims to improve the safety monitoring and emergency response capabilities of cable tunnels by integrating real-time data transmission, multi-sensor data fusion, UWB (ultra-wideband) precise positioning, and AI (Artificial Intelligence) image recognition technology.
- 4
- 4
- 4
- 4
- 4
- 4
- 4
- 4
- 4
- 4
- Harmonizing and Merging Source Models for CLIP-based Domain Generalization3
Harmonizing and Merging (HAM) is proposed, a novel source model merging framework for CLIP-based domain generalization that comprehensively consolidates source domain information while enabling mutual enhancement among source models, ultimately yielding a final model with optimal generalization capabilities.
- 3
- 3
- 3
- 3
- Why is Attention Not So Interpretable3
Theoretically analyze the combinatorial shortcuts, design one intuitive experiment to demonstrate their existence, and propose two methods to mitigate this issue, which show that the proposed methods can effectively improve the interpretability of attention mechanisms on a variety of datasets.
- 3
- 2
- 2
- Chromosome-scale genome assembly of Lepus oiostolus (Lepus, Leporidae)2
This study is at the first time to report the chromosome-scale genome for the species of the L. oiostolus, and provides a valuable genomic resource for future research on the evolution of the Leporidae.
- 2
- 2
- Combating Domain Shift with Self-Taught Labeling.2
Self-Taught Labeling (SeTL) is proposed, a new regularization approach that finds an auxiliary target-specific classifier for unlabeled data and significantly outperforms existing domain alignment techniques on a large variety of domain adaptation benchmarks.
- 2
- Multi-View Images Suffice 3D Reasoning Through Chain-of-Thought Selection and Question-Guided Fusion1
- Learning to Rank Pre-trained Vision-Language Models for Downstream Tasks1
Extensive experiments across three different benchmarks, covering a variety of application scenarios and downstream datasets, demonstrate that VEGA consistently provides reliable and accurate estimates of VLMs' performance on unlabeled downstream tasks.
- 1
- 1
- 1
- Unleashing the power of Neural Collapse for Transferability Estimation1
Fair Collapse (FaCe) is proposed for transferability estimation by comprehensively measuring the degree of neural collapse in the pre-trained model and yields state-of-the-art performance on different tasks including image classification, semantic segmentation, and text classification.
- 1
- 1
- 1
- 1
- 1
- 1
- –
- –
- Protecting Model Adaptation from Trojans in the Unlabeled Data–
This paper proposes a plug-and-play method named DiffAdapt, which can be seamlessly integrated with existing adaptation algorithms, and provides two trigger patterns with two poisoning strategies for different prior knowledge owned by attackers.
- –
- –
- –
- –
- –
- Prototypical Distillation and Debiased Tuning for Black-box Unsupervised Domain Adaptation–
Empirical results across multiple benchmarks demonstrate that ProDDing outperforms existing black-box domain adaptation methods, and in the case of hard-label black-box domain adaptation, where only predicted labels are available, ProDDing achieves significant improvements over these methods.
- –
- –
- –
- Towards Compatible Fine-tuning for Vision-Language Model Updates–
A novel approach is proposed, Class-conditioned Context Optimization (ContCoOp), which integrates learnable prompts with class embeddings using an attention layer before inputting them into the text encoder, ensuring continued effectiveness in fine-tuning.
- –
- –
- –
- –
- Self-training solutions for the ICCV 2023 GeoNet Challenge–
This work adopts a two-stage source-free domain adaptation framework with a Swin Transformer backbone to achieve knowledge transfer from the USA (source) domain to Asia (target) domain and achieves an H-score of 74.56% and ultimately ranks 1st in the GeoUniDA challenge.
- –
- –
- –
- –
- –
- –
- –
- –
- How to apply deep reinforcement learning to multi-dimensional intelligent reasoning–
Compared with knowledge representation-based, relational path-based and fusion-based methods, the experimental results show that RLPTransE method achieves better performance on large-scale dataset inference tasks.
- A construction technology of automatic reasoning system based on knowledge graph–
The proposed future knowledge reasoning model is based on the fusion knowledge representation, relational path with deep reinforcement learning methods.
- –
- –
- –
- Cross-species machine learning improves diagnostic classification of human psychiatric disorders–
An interpretable cross-species machine learning approach which extracts transgene-related core regions in the monkey brain to construct the classifier for diagnostic classification in humans, enabling improved diagnosis of autism-related disorders and dissection of their underlying circuits is reported.
- –
- –
- –
- –
- –
- –
- –
Publication data from OpenAlex, with missing venues and authors filled in from Crossref; citation counts are the higher of OpenAlex and Semantic Scholar, last synced 2026-10-14. One-sentence summaries under some papers are written by Semantic Scholar’s model. Citation counts may be lower than on Google Scholar, which indexes more sources.
Report an error
Wrong papers, two people merged into one, or a profile that should not be here? Tell us and we will fix or hide it.