Is this you? Claim this profile to correct it, add a bio and choose the work people see first.
Claim this profileAcademic lineage
Possible advisorsa guess from early papers, not confirmed
- Dacheng TaoSuggested from co-authorship
Is this you? Claim this profile to confirm or dismiss it.
Works414 from public data
- Deep Domain Generalization via Conditional Invariant Adversarial Networks848
This work proposes an end-to-end conditional invariant deep domain generalization approach by leveraging deep neural networks for domain-invariant representation learning and proves the effectiveness of the proposed method.
- 815
- CRIS: CLIP-Driven Referring Image Segmentation567
This paper designs a vision-language decoder to propagate fine-grained semantic information from textual representations to each pixel-level activation, which promotes consistency between the two modalities and presents text-to-pixel contrastive learning.
- 483
- DeepInception: Hypnotize Large Language Model to Be Jailbreaker438
A lightweight method to take advantage of the LLMs' personification capabilities to construct $\textit{a virtual, nested scene}$, allowing it to realize an adaptive way to escape the usage control in a normal scenario.
- Why ResNet Works? Residuals Generalize410
According to the obtained generalization bound, regularization terms should be introduced to control the magnitude of the norms of weight matrices not to increase too much, in practice, to ensure a good generalization ability, which justifies the technique of weight decay.
- Domain Adaptation with Conditional Transferable Components.374
This paper aims to extract conditional transferable components whose conditional distribution P(T(X)∣Y) is invariant after proper location-scale (LS) transformations, and identifies how P(Y) changes between domains simultaneously.
- Expressive power of parametrized quantum circuits337
It is proved that PQCs with a simple structure already outperform any classical neural network for generative tasks, unless the polynomial hierarchy collapses, and they are employed as an application for Bayesian learning.
- Domain Generalization via Entropy Regularization327
This work proposes an entropy regularization term that measures the dependency between the learned features and the class labels that is guaranteed to learn conditional-invariant features across all source domains and thus can learn classifiers with better generalization capabilities.
- Domain Generalization via Conditional Invariant Representations312
This paper proposes to learn a feature representation which has domain-invariant class conditional distributions P(h(X)|Y), and proposes a conditional invariant representation which can be guaranteed if the class prior P(Y) does not change across training and test domains.
- Multiple Kernel k-means with Incomplete Kernels306
Two simple yet effective algorithms that perform multiple kernel clustering directly with the presence of incomplete kernel matrices, which are treated as auxiliary variables to be jointly optimized and which consistently achieve superior performance are proposed.
- Sub-center ArcFace: Boosting Face Recognition by Large-Scale Noisy Web Faces305
The intra-class constraint of ArcFace is relaxed to improve the robustness to label noise and recapture intra-compactness, decrease the influence from noise, and achieve comparable performance compared to ArcFace trained on the manually cleaned dataset.
- Dual T: Reducing Estimation Error for Transition Matrix in Label-noise Learning298
This paper introduces an intermediate class to avoid directly estimating the noisy class posterior of the transition matrix, and introduces the dual $T-estimator for estimating transition matrices, leading to better classification performances.
- Fast Supervised Discrete Hashing283
This paper proposes a new learning-based hashing method called "fast supervised discrete hashing" (FSDH) based on “supervised discrete hashing” (SDH), which uses a very simple yet effective regression of the class labels of training examples to the corresponding hash code to accelerate the algorithm.
- Selective-Supervised Contrastive Learning with Noisy Labels258
Experiments on multiple noisy datasets demonstrate the robustness of the learned representations by the proposed selective-supervised contrastive learning (Sel-CL), following the state-of-the-art performance.
- Control Batch Size and Learning Rate to Generalize Well: Theoretical and Empirical Evidence247
A PAC-Bayes generalization bound for neural networks trained by SGD is proved, which has a positive correlation with the ratio of batch size to learning rate, which builds the theoretical foundation of the training strategy.
- Learning with Biased Complementary Labels224
A framework that provides a general method to modify traditional loss functions and extends standard deep neural network classifiers to learn with biased complementary labels and theoretically ensures that the classifier learned with complementary labels converges to the optimal one learned with true labels.
- 204
- 187
- Quantum noise protects quantum classifiers against adversaries182
By taking advantage of depolarisation noise in quantum circuits for classification, a robustness bound against adversaries can be derived where the robustness improves with increasing noise, which is the first quantum protocol that can be used against the most general adversaries.
- Semantic Structure-based Unsupervised Deep Hashing176
This work designs a deep architecture and a pair-wise loss function to preserve the semantic structure of the semantic relationships between points in unsupervised settings, and shows that SSDH significantly outperforms current state-of-the-art methods.
- 173
- Orthogonal Deep Neural Networks169
This paper proves that DNNs are of local isometry on data distributions of practical interest, and establishes a new generalization error bound that is both scale- and range-sensitive to singular value spectrum of each of networks’ weight matrices.
- DistillHash: Unsupervised Deep Hashing by Distilling Data Pairs169
A new deep unsupervised hashing model, called DistilHash, is proposed, which can learn a distilled data set, where data pairs have confident similarity signals and the semantic similarity labels assigned by the optimal Bayesian classifier can be potentially distilled.
- FedFed: Feature Distillation against Data Heterogeneity in Federated Learning167
This work proposes a novel approach called FedFed, which partitions data into performance-sensitive features and performance-robust features that are globally shared to mitigate data heterogeneity, while the performance-Robust features are kept locally.
- Multiview Matrix Completion for Multilabel Image Classification160
Experimental evaluation on two real-world data sets demonstrate the effectiveness of MVMC for transductive (semisupervised) multilabel image classification, and show that MVMC can exploit complementary properties of different features and output-consistent labels for improved multilabe image classification.
- 158
- A Second-Order Approach to Learning with Instance-Dependent Label Noise157
This work proposes and studies the potentials of a second-order approach that leverages the estimation of several covariance terms defined between the instance-dependent noise rates and the Bayes optimal label, and shows that this set of second- order statistics successfully captures the induced imbalances.
- Spectral Ensemble Clustering154
Experimental results on various real-world data sets demonstrate that SEC is an effective and efficient competitor to some state-of-the-art ensemble clustering methods and is also suitable for big data clustering.
- A Regularization Approach for Instance-Based Superset Label Learning146
A novel regularization approach for instance-based superset label (RegISL) learning is developed so that the instance- based method also inherits the good discriminative ability possessed by the regularization scheme.
- 144
- Recent Advances for Quantum Neural Networks in Generative Learning143
This paper interprets these QGLMs, covering quantum circuit Born machines, quantum generative adversarial networks, quantum Boltzmann machines, and quantum variational autoencoders, as the quantum extension of classical generative learning models, and explores their intrinsic relations and their fundamental differences.
- Algorithm-Dependent Generalization Bounds for Multi-Task Learning133
This paper presents novel algorithm-dependent generalization bounds for MTL by exploiting the notion of algorithmic stability and shows that the similarity of feature structures in MTL will lead to specific regularizations for predicting, which enables the learning algorithms to generalize fast and correctly from a few examples.
- DeepSolo: Let Transformer Decoder with Explicit Points Solo for Text Spotting128
DeepSolo, a simple DETR-like baseline that lets a single Decoder with Explicit Points Solo for text detection and recognition simultaneously, and introduces a text-matching criterion to deliver more accurate supervisory signals, thus enabling more efficient training.
- Killing Two Birds with One Stone: Efficient and Robust Training of Face Recognition CNNs by Partial FC128
A sparsely updating variant of the FC layer, named Partial FC (PFC), where all class centers are still maintained throughout the whole training process, but only a subset is selected and updated in each iteration, resulting in a dramatically reduced computing requirement and frequency of passive update on tail class centers.
- 128
- 126
- 126
- Unveiling Causal Reasoning in Large Language Models: Reality or Mirage?124
This work proposes G^2-Reasoner, a method that incorporates general knowledge and goal-oriented prompts into LLMs'causal reasoning processes, and demonstrates that G^2-Reasoner significantly enhances LLMs'causal reasoning capability, particularly in fresh and counterfactual contexts.
- HumanMAC: Masked Motion Completion for Human Motion Prediction124
The proposed framework enjoys promising algorithmic properties, which only needs one loss in optimization and is trained in an end-to-end manner and accomplishes the switch of different categories of motions effectively, which is significant in realistic tasks, e.g., the animation task.
- 124
- 121
- 120
- On the learnability of quantum neural networks120
This work derives the convergence performance of QNN under the NISQ setting, and identifies classes of computationally hard concepts that can be efficiently learned by QNN, and proves that any concept class, which is efficiently learnable by a restricted quantum statistical query (QSQ) learning model, can also be efficiently learning byQNN.
- Decomposition-Based Transfer Distance Metric Learning for Image Classification118
This paper considers the transfer learning setting by exploiting the large quantity of side information from certain related, but different source tasks to help with target metric learning (with only a little side information), and proposes a decomposition-based transfer DML (DTDML).
- Point-Query Quadtree for Crowd Counting, Localization, and More110
For the first time, a single counting model can address multiple crowd-related tasks across different learning paradigms and is implemented via data-dependent quadtree, where each querying point could split into four new points when necessary, thus enabling dynamic processing of sparse and dense regions.
- Truncated Cauchy Non-Negative Matrix Factorization110
A Truncated CauchyNMF loss that handle outliers by truncating large errors, and developed to robustly learn the subspace on noisy datasets contaminated by outliers are proposed.
- Correcting the Triplet Selection Bias for Triplet Loss107
A new variant of triplet loss is proposed, which tries to reduce the bias in triplet selection by adaptively correcting the distribution shift on the selected triplets.
- Supervised Discrete Hashing With Relaxation102
Compared with SDH, which uses the traditional zero-one matrix, SDHR utilizes the learned regression target matrix and, therefore, more accurately measures the classification error of the regression model and is more flexible.
- Instance-Dependent Label-Noise Learning with Manifold-Regularized Transition Matrix Estimation100
The proposed manifold-regularized technique works by directly reducing the estimation error without hurting the approximation error about the estimation problem of T(x), and demonstrates that the method is superior to state-of-the-art approaches for label-noise learning under the challenging IDN.
- ALIP: Adaptive Language-Image Pre-training with Synthetic Caption99
An Adaptive Language-Image Pre-training (ALIP), a bi-path model that integrates supervision from both raw text and synthetic caption that achieves state-of-the-art performance on multiple downstream tasks including zero-shot image-text retrieval and linear probe is proposed.
- Transferring Knowledge Fragments for Learning Distance Metric from a Heterogeneous Domain99
This paper examines the problem of transfer distance metric learning (DML), which usually aims to mitigate the label information deficiency issue in the target DML, and develops novel algorithm to learn either linear or nonlinear target metric.
- 97
- Negative Label Guided OOD Detection with Pretrained Vision-Language Models95
This paper proposes a novel post hoc OOD detection method, called NegLabel, which takes a vast number of negative labels from extensive corpus databases, and designs a novel scheme for the OOD score collaborated with negative labels.
- 91
- Understanding Robust Overfitting of Adversarial Training and Beyond90
The causes of robust overfitting are explored by comparing the data distribution of non-overfit and strong adversary adversarial training, and it is identified that some small-loss data which are not worthy of the adversary strength cause robust over fitting in the strong adversary mode.
- 90
- 89
- 85
- 84
- 84
- Local Rademacher Complexity for Multi-Label Learning82
This work analyzes the local Rademacher complexity of empirical risk minimization-based multi-label learning algorithms, and proposes a new conditional singular value thresholding algorithm that has a sharper generalization error bound.
- Towards Label-free Scene Understanding by Vision Foundation Models81
This paper proposes a novel Cross-modality Noisy Supervision (CNS) method that leverages the strengths of CLIP and SAM to supervise 2D and 3D networks simultaneously, and introduces a prediction consistency regularization to co-train 2Dand3D networks, then further impose the networks' latent space consistency using the SAM's robust feature representation.
- 81
- Continuous Dropout.78
The proposed continuous dropout is considerably closer to the activation characteristics of neurons in the human brain than traditional binary dropout and has the property of avoiding the co-adaptation of feature detectors, which suggests that it can extract more independent feature detectors for model averaging in the test stage.
- Safety at Scale: A Comprehensive Survey of Large Model and Agent Safety76
This survey provides a systematic review of current safety research on large models, covering Vision Foundation Models, Large Language Models, Vision-Language Pre-training models, Vision-Language Models, Diffusion Models, and large-model-powered Agents, and presents a comprehensive taxonomy of safety threats to these models.
- BiCro: Noisy Correspondence Rectification for Multi-modality Data via Bi-directional Cross-modal Similarity Consistency74
The experiments on three popular cross-modal matching datasets demonstrate that the proposed BiCro method significantly improves the noise-robustness of various matching models, and surpass the state-of-the-art by a clear margin.
- Bridging the Gap Between Few-Shot and Many-Shot Learning via Distribution Calibration72
A generalization error bound is established for the proposed distribution-calibration-based few-shot learning, which consists of the distribution assumption error, the distribution approximation error, and the estimation error that theoretically justifies the effectiveness of the proposed method.
- 72
- Harnessing Out-Of-Distribution Examples via Augmenting Content and Style70
A HOOD method that can leverage the content and style from each image instance to identify benign and malign OOD data is proposed, and the proposed novel disentanglement and data augmentation techniques can effectively deal with OOD examples in unknown and open environments.
- 67
- Out-of-distribution Detection with Implicit Outlier Transformation65
This work proposes a novel OE-based approach that makes the model perform well for unseen OOD situations, even forseen OOD cases, and offers an efficient way of synthesizing OOD data, which can further benefit the detection model, besides the surrogate Ood data.
- Parts-dependent Label Noise: Towards Instance-dependent Label Noise65
Empirical evaluations demonstrate this method is superior to the state-of-the-art approaches for learning from the instance-dependent label noise and the transition matrices for parts of the instance are learned by exploiting anchor points.
- 64
- 64
- 61
- 61
- Unicom: Universal and Compact Representation Learning for Image Retrieval60
This paper clusters the large-scale LAION400M into one million pseudo classes based on the joint textual and visual features extracted by the CLIP model and randomly selects partial inter-class prototypes to construct the margin-based softmax loss, making the classification conflict-robust and the feature embedding compact.
- 60
- An Efficient and Provable Approach for Mixture Proportion Estimation Using Linear Independence Assumption59
This paper proposes a method that uniquely identifies the mixture proportions, whose output provably converges to the optimal solution, and that is computationally efficient and shows the superiority of the proposed method over the state-of-the-art methods in two applications including learning with label noise and semi-supervised learning on both synthetic and real-world datasets.
- Dimensionality-Dependent Generalization Bounds for k-Dimensional Coding Schemes58
A dimensionality-dependent generalization bound is derived for k-dimensional coding schemes by bounding the covering number of the loss function class induced by the reconstruction error, and it is shown that this bound can be tighter than previous results because it avoids inducing the worst-case upper bound on k of theloss function.
- Positive and Unlabeled Learning with Label Disambiguation57
A novel algorithm dubbed as “Positive and Unlabeled learning with Label Disambiguation” (PULD) is proposed, which first regard all the unlabeled examples in PU learning as ambiguously labeled as positive and negative, and then employs the margin-based label disambIGuation strategy, which enlarges the margin of classifier response between the most likely label and the less likely one, to find the unique ground-truth label of each unlabeling example.
- Relational Subsets Knowledge Distillation for Long-Tailed Retinal Diseases Recognition56
This study proposes class subset learning by dividing the long-tailed data into multiple class subsets according to prior knowledge, such as regions and phenotype information, and enforces the model to focus on learning the subset-specific knowledge.
- Enhancing One-Shot Federated Learning Through Data and Ensemble Co-Boosting55
This work introduces a novel framework, Co-Boosting, in which synthesized data and the ensemble model mutually enhance each other progressively, and leverages the current ensemble model to synthesize higher-quality samples in an adversarial manner.
- Diversified Outlier Exposure for Out-of-Distribution Detection via Informative Extrapolation55
This work proposes a novel framework, namely, Diversified Outlier Exposure (DivOE), for effective OOD detection via informative extrapolation based on the given auxiliary outliers by explicitly synthesizing more informative outliers for extrapolation during training.
- 55
- Bilateral Dependency Optimization54
Experiments show that BiDO achieves the state-of-the-art defense performance for a variety of datasets, classifiers, and MI attacks while suffering a minor classification-accuracy drop compared to the well-trained classifier with no defense, which lights up a novel road to defend against MI attacks.
- Out-of-distribution Detection Learning with Unreliable Out-of-distribution Sources53
A powerful data generation-based learning method named Auxiliary Task-based OOD Learning (ATOL) is proposed that can relieve the mistaken OOD generation and is conducted under various OOD detection setups.
- 53
- FlatMatch: Bridging Labeled Data and Unlabeled Data with Cross-Sharpness for Semi-Supervised Learning52
FlatMatch is proposed which minimizes a cross-sharpness measure to ensure consistent learning performance between the two datasets and mitigates the mismatched learning performance, enabling the effective exploitation of unlabeled data and improving SSL performance.
- Early Stopping Against Label Noise Without Validation Data51
A novel early stopping method called Label Wave is proposed, which does not require validation data for selecting the desired model in the presence of label noise and works by tracking the changes in the model's predictions on the training set during the training process, aiming to halt training before the model unduly fits mislabeled data.
- Exploring Set Similarity for Dense Self-supervised Representation Learning50
This paper proposes to explore set similarity (SetSim) for dense self-supervised representation learning, and generalizes pixel-wise similarity learning to set-wise one to improve the robustness because sets contain more semantic and structure information.
- BadLabel: A Robust Perspective on Evaluating and Enhancing Label-Noise Learning49
A robust LNL method is proposed that perturbs the labels in an adversarial manner at each epoch to make the loss values of clean and noisy labels again distinguishable and effectively improve the generalization performance of the model under various types of label noise.
- A Grover-search based quantum learning scheme for classification49
A Grover-search based quantum learning scheme (GBLS) is devised to address the above two issues and provides guidance to develop advanced quantum classifiers on near-term quantum devices and opens up an avenue to explore potential quantum advantages in various classification tasks.
- 48
- On Better Exploring and Exploiting Task Relationships in Multitask Learning: Joint Model and Feature Learning48
This paper proposes a new strategy to measure the relatedness that jointly learns shared parameters and shared feature representations and conducts various experiments to verify the superiority of the proposed joint model and feature MTL method.
- 48
- 47
- 47
- Exploring Criteria of Loss Reweighting to Enhance LLM Unlearning46
Empirical results on extensive datasets validate the efficacy of the proposed SatImp, a simple reweighting method that combines the advantages of both saturation and importance, potentially bridging existing research gaps and indicating directions for future research.
- Robust Generalization Against Photon-Limited Corruptions via Worst-Case Sharpness Minimization46
This work proposes SharpDRO, which exhibits a strong generalization ability against severe corruptions and exceeds well-known baseline methods with large performance gains by penalizing the sharpness of the worst-case distribution, which measures the loss changes around the neighbor of learning parameters.
- 45
- LTF: A Label Transformation Framework for Correcting Label Shift45
This paper proposes an end-to-end Label Transformation Framework (LTF) for correcting target shift, which implicitly models the shift of P Y and the conditional distribution P X | Y using neural networks.
- Generative-Discriminative Complementary Learning45
A generative-discriminative complementary learning method that estimates the ordinary labels by modeling both the conditional and instance distributions, which improves the accuracy of predicting ordinary labels and is able to generate high-quality instances in spite of weak supervision.
- Reliable Multi-View Clustering45
Experimental results on a number of data sets demonstrate that the proposed Reliable Multi-View Clustering method can effectively improve the reliability of multi-view clustering.
- Watermarking for Out-of-distribution Detection44
A unified pattern that is superimposed onto features of original data is learned, and the model's detection capability is largely boosted after watermarking, demonstrating the significance of the reprogramming property of deep models in OOD detection.
- Reinforcement Learning with Verifiable yet Noisy Rewards under Imperfect Verifiers42
Two lightweight corrections are implemented as lightweight hooks in a group relative policy optimization pipeline that improve RLVR for math reasoning under synthetic and real verifier noise, with the forward variant being more stable under heavier noise.
- Envisioning Outlier Exposure by Large Language Models for Out-of-Distribution Detection42
This paper proposes to tackle the constraint of expert knowledge and reasoning capability of large language models to Envision potential Outlier Exposure, termed EOE, without access to any actual OOD data by leveraging the expert knowledge and reasoning capability of large language models to Envision potential Outlier Exposure, termed EOE.
- Evolving Semantic Prototype Improves Generative Zero-Shot Learning42
A dynamic semantic prototype evolving (DSP) method is proposed to align the empirically predefined semantic prototypes and the real prototypes for class-related feature synthesis and indicates that evolving semantic prototype explores a virgin field in ZSL.
- Quantum Differentially Private Sparse Regression Learning42
An efficient quantum differentially private (QDP) Lasso estimator to solve sparse regression tasks and implies that the QDP Lasso can be faster than both the optimal classical and quantum non-private Lasso.
- Understanding How Feature Structure Transfers in Transfer Learning42
This paper proposes a general analysis scheme to theoretically justify that if the source and target domains share similar feature structures, the source domain feature structure is transferable to the target domain, regardless of the change of the labeling functions across domains.
- Joint Admission Control and Resource Allocation of Virtual Network Embedding via Hierarchical Deep Reinforcement Learning41
A deep Hierarchical Reinforcement Learning approach to learn a joint Admission Control and Resource Allocation policy for VNE, named HRL-ACRA, and develops a deep feature-aware graph neural network to capture the features of VNR and physical network and exploit a sequence-to-sequence model to generate embedding actions iteratively.
- Removing Adversarial Noise in Class Activation Feature Space41
Empirical evaluations demonstrate that the proposed self-supervised adversarial training mechanism could significantly enhance adversarial robustness in comparison to previous state-of-the-art approaches, especially against unseen adversarial attacks and adaptive attacks.
- Label-Noise Robust Domain Adaptation.41
This paper is the first to comprehensively investigate how label noise could adversely affect existing domain adaptation methods in various scenarios and theoretically prove that there exists a method that can essentially reduce the side-effect of noisy source labels in domain adaptation.
- Trustable Co-Label Learning From Multiple Noisy Annotators39
This paper proposes a data-efficient approach, called Trustable Co-label Learning (TCL), to learn deep classifiers from multiple noisy annotators when a small set of trusted data is available, which effectively uses trusted data as a guide to generate trustable soft labels (termed co-labels).
- Revisiting Knowledge Distillation: An Inheritance and Exploration Framework39
This paper proposes a novel inheritance and exploration knowledge distillation framework (IE-KD), in which a student model is split into two parts - inheritance and Exploration, which can be a general technique to improve the student network to achieve SOTA performance.
- DEEM: Diffusion Models Serve as the Eyes of Large Language Models for Image Perception38
Compared to the state-of-the-art interleaved content generation models, DEEM exhibits enhanced robustness and a superior capacity to alleviate model hallucinations while utilizing fewer trainable parameters, less pre-training data, and a smaller base model size.
- SimT: Handling Open-set Noise for Domain Adaptive Semantic Segmentation38
A simplex noise transition matrix (SimT) is proposed to model the mixed noise distributions in DA semantic segmentation and formulate the problem as estimation of SimT to correct noise issues in pseudo labels and promote the generalization ability of segmentation model on target domain data.
- 38
- 37
- LaVin-DiT: Large Vision Diffusion Transformer36
This work introduces a novel pathway for large vision foundation models, underscoring the promising potential of diffusion transformers, and introduces LaVin-DiT, a scalable and unified foundation model designed to tackle over 20 computer vision tasks in a generative framework.
- Machine Vision Therapy: Multimodal Large Language Models Can Enhance Visual Robustness via Denoising In-Context Learning36
This paper proposes a novel Denoising In-Context Learning (DICL) strategy to align vision tasks with MLLMs and proposes an instruction containing a correct exemplar and an erroneous one from the most probable noisy class can be constructed to rectify the noisy predictions from vision models.
- MissDAG: Causal Discovery in the Presence of Missing Data with Continuous Additive Noise Models36
MissDAG maximizes the expected likelihood of the visible part of observations under the expectation-maximization (EM) framework and uses a likelihood-based causal discovery algorithm with directed acyclic graph constraint for incorporating various causal discovery algorithms.
- 36
- Multi-task model and feature joint learning36
A theoretical bound is presented which directly demonstrates that the proposed multi-task learning method can successfully model the relatedness via joint common parameter- and common feature-learning.
- 35
- Tackling Instance-Dependent Label Noise via a Universal Probabilistic Model34
This paper proposes a simple yet universal probabilistic model, which explicitly relates noisy labels to their instances, and can be realized by DNNs, where the training procedure is accomplished by employing a novel alternating optimization algorithm.
- From Debate to Equilibrium: Belief-Driven Multi-Agent LLM Reasoning via Bayesian Nash Equilibrium33
This work introduces Efficient Coordination via Nash Equilibrium (ECON), a hierarchical reinforcement-learning paradigm that marries distributed reasoning with centralized final output and mathematically proves that ECON attains a markedly tighter regret bound than non-equilibrium multi-agent schemes.
- Quantization Aware Attack: Enhancing Transferable Adversarial Attacks by Model Quantization33
This work proposes quantization aware attack (QAA), which fine-tunes a QNN substitute model with a multiple-bitwidth training objective and demonstrates that QAA addresses the two issues that are commonly known to hinder transferability: 1) quantization shifts and 2) gradient misalignments.
- 33
- Counterfactual Fairness with Partially Known Causal Graph32
Interestingly, it is found that counterfactual fairness can be achieved as if the true causal graph were fully known, when specific background knowledge is provided: the sensitive attributes do not have ancestors in the causal graph.
- Symmetric Pruning in Quantum Neural Networks31
This work proposes the effective quantum neural tangent kernel (EQNTK) and connects this concept with over-parameterization theory to quantify the convergence of QNNs towards the global optima and devise a symmetric pruning scheme to greatly improve the performance ofQNNs when the explicit symmetry information of Hamiltonian is unavailable.
- 31
- COVID-MTL: Multitask learning with Shift3D and random-weighted loss for COVID-19 diagnosis and severity assessment31
An end-to-end multitask learning (MTL) framework that is capable of automated and simultaneous detection (against both radiology and NAT) and severity assessment of COVID-19 and is able to accelerate convergence and improve joint learning performance compared to single-task models is presented.
- How Well Does GPT-4V(ision) Adapt to Distribution Shifts? A Preliminary Investigation30
This study rigorously evaluates GPT-4V's adaptability and generalization capabilities in dynamic environments, benchmarking against prominent models like CLIP, LLaVA, and Gemini and delves into GPT-4V's zero-shot generalization across 13 diverse datasets spanning natural, medical, and molecular domains.
- 30
- Model Inversion Attacks: A Survey of Approaches and Countermeasures29
A threat-model- and assumption-aware synthesis of attacks and defenses is provided, which compares exposed interfaces, attacker knowledge, reconstruction priors and targets, failure modes, deployment constraints, and privacy-utility trade-offs, while highlighting modeling principles, optimization challenges, and future directions.
- Late Stopping: Avoiding Confidently Learning from Mislabeled Examples29
This paper proposes a new framework, Late Stopping, which leverages the intrinsic robust learning ability of DNNs through a prolonged training process, and empirically observes that mislabeled and clean examples exhibit differences in the number of epochs required for them to be consistently and correctly classified, and thus high-probability mislabeled examples can be removed.
- IDEAL: Influence-Driven Selective Annotations Empower In-Context Learners in Large Language Models29
An influence-driven selective annotation method that aims to minimize annotation costs while improving the quality of in-context examples, which works in an end-to-end manner, avoids an intractable explicit balance between data diversity and representativeness, and enjoys theoretical support.
- 29
- Transferring Annotator- and Instance-Dependent Transition Matrix for Learning From Crowds28
Theoretical analyses are derived to demonstrate that both the knowledge transfer from global to individuals and the knowledge transfer between neighboring individuals can effectively help mitigate the challenge of modeling general AIDTM.
- FedImpro: Measuring and Improving Client Update in Federated Learning28
FedImpro decouples the model into high-level and low-level components, and trains the high-level portion on reconstructed feature distributions that enhances the generalization contribution and reduces the dissimilarity of gradients in FL.
- Extended : Learning with Mixed Closed-set and Open-set Noisy Labels28
The traditional transition matrix is extended to be able to model mixed label noise, and further to the cluster-dependent extended transition matrix to better combat the instance-dependent label noise in real-world applications.
- 28
- 28
- NoiseGPT: Label Noise Detection and Rectification through Probability Curvature27
NoiseGPT is proposed to effectively leverage MLLMs as a knowledge expert for conducting label noise detection and rectification and observes a probability curvature effect of MLLMs where clean and noisy examples reside on curvatures with different smoothness, further enabling the detection of label noise.
- QUBIQ: Uncertainty Quantification for Biomedical Image Segmentation Challenge27
The obtained results indicate the importance of the ensemble models, as well as the need for further research to develop efficient 3D methods for uncertainty quantification methods in 3D segmentation tasks.
- InstanT: Semi-supervised Learning with Instance-dependent Thresholds27
This paper devise a novel instance-dependent threshold function for all unlabeled instances by utilizing their instance-level ambiguity and the instance- dependent error rates of pseudo-labels, so instances that are more likely to have incorrect pseudo-Labels will have higher thresholds.
- 27
- 27
- Tackling Noisy Labels With Network Parameter Additive Decomposition26
To decouple the memorization of clean data and mislabeled data, and further reduce the side effect of mislabeled data, additive decomposition on network parameters is performed.
- Eliminating Catastrophic Overfitting Via Abnormal Adversarial Examples Regularization26
A novel method is designed, termed Abnormal Adversarial Examples Regularization (AAER), which explicitly regularizes the variation of AAEs to hinder the classifier from becoming distorted, and can effectively eliminate CO and further boost adversarial robustness with negligible additional computational overhead.
- Discovering and Reasoning of Causality in the Hidden World with Large Language Models26
COAT incorporates LLMs as an factor proposer that extracts the potential causal factors from unstructured data and verifies the effectiveness of COAT in uncovering the underlying causal system with two case studies of review rating analysis and neuropathic diagnosis.
- Regularly Truncated M-Estimators for Learning With Noisy Labels26
Regular truncated M-estimators (RTME) is proposed, which can alternately switch modes between truncated M-estimators and original M-estimators and is robust to broad noise types and levels.
- Improving Adversarial Robustness via Mutual Information Estimation26
This paper investigates the dependence between outputs of the target model and input adversarial samples from the perspective of information theory, and proposes an adversarial defense method that could effectively improve the adversarial accuracy against multiple attacks.
- 26
- 26
- 26
- AI-Accelerated Discovery of Electrocatalyst Materials25
This review highlights the pivotal role of AI in electrocatalyst discovery, focusing on the critical aspects of data, descriptors, and machine learning models, and discusses various AI approaches, including their applications in accelerating DFT calculations, exploring reaction mechanisms, designing electrocatalysts, and predicting performance.
- Federated Causal Discovery from Heterogeneous Data25
A novel FCD method attempting to accommodate arbitrary causal models and heterogeneous data by utilizing a surrogate variable corresponding to the client index to account for the data heterogeneity across different clients and establishing a federated independent change principle to determine causal directions.
- Beyond Gaussians: Fast and High-Fidelity 3D Splatting with Linear Kernels25
3D Linear Splatting (3DLS), which replaces Gaussian kernels with linear kernels to achieve sharper and more precise results, particularly in high-frequency regions, demonstrates state-of-the-art fidelity and accuracy.
- ERASE: Error-Resilient Representation Learning on Graphs for Label Noise Tolerance25
The proposed ERASE (Error-Resilient representation learning on graphs for lAbel noiSe tolerancE) allows us to more effectively withstand errors caused by mislabeled nodes, thereby strengthening the robustness of deep networks in handling noisy graph data.
- On the Over-Memorization During Natural, Robust and Catastrophic Overfitting25
This work examines the memorization effect in DNNs and reveals a shared behaviour termed over-memorization, which impairs their generalization capacity, and proposes a general framework, Distraction Over-Memorization (DOM), which explicitly prevents over-memorization by either removing or augmenting the high-confidence natural patterns.
- 24
- 24
- Learning the Latent Causal Structure for Modeling Label Noise22
By learning the latent causal structure governing the generating process of noisy data, this method can estimate noise transition matrices without the need for similarity-based assumptions and model them with a learnable graphical model.
- Unleashing Mask: Explore the Intrinsic Out-of-Distribution Detection Capability22
A novel method is proposed, Unleashing Mask, which aims to restore the OOD discriminative capabilities of the well-trained model with ID data by utilizing a mask to figure out the memorized atypical samples, and then finetune the model or prune it with the introduced mask to forget them.
- 22
- 22
- Learning General and Specific Embedding with Transformer for Few-Shot Object Detection21
This paper proposes a Transformer-based general and specific embedding learning (T-GSEL) method for FSOD, which achieves compelling FSOD performance on both PASCAL VOC and MS COCO datasets compared with other state-of-the-art approaches.
- 21
- 21
- 21
- 21
- 21
- Winning Prize Comes from Losing Tickets: Improve Invariant Learning by Exploring Variant Parameters for Out-of-Distribution Generalization20
By fully exploring both variant and invariant parameters, the EVIL can effectively identify a robust subnetwork to improve OOD generalization and can effectively and efficiently enhance many popular methods, such as ERM, IRM, SAM, etc.
- 20
- Dual-Path Distillation: A Unified Framework to Improve Black-Box Attacks20
A novel framework, dual-path distillation, is proposed that utilizes the feedback knowledge not only to craft adversarial examples but also to alter the searching directions to achieve efficient attacks.
- 20
- 20
- Open-Vocabulary Segmentation with Unpaired Mask-Text Supervision19
A novel weakly-supervised open-vocabulary segmentation framework that learns from unpaired image-mask and image-text pairs, which can be independently and efficiently collected, and designs a multi-scale matching strategy to reduce noisy mask-entity pairs.
- 19
- 19
- Implementable Quantum Classifier for Nonlinear Data19
The main ingredients of the method are variational quantum perceptron (VQP) and a quantum generalization of classical ensemble learning, which employs parameterized quantum circuits to learn a Grover search operation with classical optimization, and can achieve quadratic speedup in query complexity compared to its classical counterparts.
- E2HQV: High-Quality Video Generation from Event Camera via Theory-Inspired Model-Aided Deep Learning18
- PADDLES: Phase-Amplitude Spectrum Disentangled Early Stopping for Learning with Noisy Labels18
This work proposes early stops at different times for AS and PS by disentangling the features of some layer(s) into AS and PS using Discrete Fourier Transform (DFT) during training, and shows that PS can increase the robustness of CNNs to label noise, more so than AS can.
- 18
- Why Controlling the Asymptomatic Infection Is Important: A Modelling Study with Stability and Sensitivity Analysis18
A six-dimensional ODE system (SEIAQR epidemical model) is established to study the dynamics of COVID-19 spreading considering infection by exposed, infected, and asymptomatic cases and it is found that the asymPTomatic infection greatly affected the endemic equilibrium.
- Type-I Generative Adversarial Attack17
GAA aims at exploiting the distribution mapping from the source domain of multiple classes to the target domain of a single class by using generative adversarial networks to exploit the ambiguous distributions among different classes in the resultant feature space of the model.
- 17
- Label Distribution Learning with Biased Annotations Assisted by Multi-Label Learning16
This paper addresses the LDL with biased annotations problem from a novel perspective, where it first degenerates the soft label distribution into a hard multi-hot label and then recovers the true label information for each instance.
- Noisy Test-Time Adaptation in Vision-Language Models16
The Adaptive Noise Detector (AdaND), which utilizes the frozen model's outputs as pseudo-labels to train a noise detector, and injects Gaussian noise during adaptation, preventing the detector from misclassifying clean samples as noisy.
- NoiseDiffusion: Correcting Noise for Image Interpolation with Diffusion Models beyond Spherical Linear Interpolation16
NoiseDiffusion performs interpolation within the noisy image space and injects raw images into these noisy counterparts to address the challenge of information loss and enables us to interpolate natural images without causing artifacts or information loss, thus achieving the best interpolation results.
- Neural auto-designer for enhanced quantum kernels16
This study presents a data-driven approach that automates the design of problem-specific quantum feature maps and incorporates a deep neural predictor to efficiently evaluate the performance of various candidate quantum kernels.
- Multiscale Representation for Real-Time Anti-Aliasing Neural Rendering16
This work proposes mip voxel grids (Mip-VoG), an explicit multiscale representation with a deferred architecture for real-time anti-aliasing rendering simultaneously, and is the first to offer multiscale training and real-time anti-aliasing rendering simultaneously.
- Meta Clustering Learning for Large-scale Unsupervised Person Re-identification16
This paper makes the first attempt to the large-scale U-ReID and proposes a "small data for big task" paradigm dubbed Meta Clustering Learning (MCL), which only pseudo-labels a subset of the entire unlabeled data via clustering to save computing for the first-phase training.
- 16
- 16
- 16
- 16
- Online Heterogeneous Transfer Metric Learning16
The proposed online HTML (OHTML) method formulate the problem in the online setting so that the optimization is efficient and the model can be adapted to new coming data and demonstrates both effectiveness and efficiency of the proposed method.
- Mitigating Label Noise on Graph via Topological Sample Selection15
A TSS method is proposed that boosts the informative sample selection process in a graph by utilising topological information and theoretically proves that the procedure minimizes an upper bound of the expected risk under target clean distribution, and experimentally shows the superiority of the method compared with state-of-the-art baselines.
- 15
- 15
- 15
- 15
- 15
- 15
- 15
- 15
- Maximum Mean Discrepancy is Aware of Adversarial Attacks15
It is validated that MMD is aware of adversarial attacks, which lights up a novel road for adversarial attack detection based on two-sample tests.
- 14
- Robust Training of Federated Models with Extremely Label Deficiency14
This work proposes a novel twin-model paradigm, called Twin-sight, designed to enhance mutual guidance by providing insights from different perspectives of labeled and unlabeled data, and introduces a neighbourhood-preserving constraint that encourages the preservation of the neighbourhood relationship among data features extracted by both models.
- 14
- Adversarial Training with Complementary Labels: On the Benefit of Gradually Informative Attacks14
A new learning strategy using gradually informative attacks, which consists of two critical components: Warm-up Attack (Warm-up) gently raises the adversarial perturbation budgets to ease the adversaria optimization with CLs; Pseudo-Label Attack (PLA) incorporates the progressively informative model predictions into a corrected complementary loss.
- 14
- 14
- 14
- 14
- 14
- Quantum-inspired algorithm for general minimum conical hull problems14
This paper proposes a sublinear classical algorithm to tackle general minimum conical hull problems when the input has stored in a sample-based low-overhead data structure and achieves the exponential speedup over DCA and, therefore, provides advantages for high dimensional problems.
- Decomposed Prompt Decision Transformer for Efficient Unseen Task Generalization13
This paper proposes the Decomposed Prompt Decision Transformer (DPDT), a two-stage paradigm to efficiently learn prompts for unseen tasks in a parameter-efficient manner that incorporates parameters from pre-trained language models (PLMs) to initialize DPDT, thereby providing rich prior knowledge encoded in language models.
- Pseudo-Private Data Guided Model Inversion Attacks13
This paper proposes increasing the density around high-quality pseudo-private data—recovered samples through model inversion that exhibit characteristics of the private training data—by slightly tuning the generator, which effectively increases the probability of sampling actual private data that is close to these pseudo-private data during the inversion process.
- Architecture, Dataset and Model-Scale Agnostic Data-free Meta-Learning13
A unified framework, dubbed PURER, is proposed, which contains a unified framework for meta training with ECI as an adversarial form in an end-to-end manner and a simple plug-and-play supplement—ICFIL—only used during meta testing to narrow the gap between meta training and meta testing task distribution.
- Quantum Divide-and-Conquer Anchoring for Separable Non-negative Matrix Factorization13
This paper investigates how the power of quantum computation can be capitalized to solve the non-negative matrix factorization with the separability assumption (SNMF) by devising a quantum algorithm based on the divide-and-conquer anchoring (DCA) scheme.
- Label Distribution Learning with Biased Annotations by Learning Multi-Label Representation12
This paper addresses the LDL with biased annotations problem from a novel perspective, where it first degenerates the soft label distribution into a hard multi-hot label and then recovers the true label information for each instance.
- Prompt-Based Multi-Interest Learning Method for Sequential Recommendation12
A prompt-based multi-interest learning method (PoMRec), where specific prompts are inserted into the inputted user interactions to make them adaptive to the multi-interest extractor and aggregator, which outperforms the state-of-the-art multi-interest learning methods.
- 12
- HCVP: Leveraging Hierarchical Contrastive Visual Prompt for Domain Generalization12
This work introduces the novel Hierarchical Contrastive Visual Prompt (HCVP) methodology, which represents a significant advancement in the field, setting itself apart with a unique generative approach to prompts, alongside an explicit model structure and specialized loss functions.
- Layer-Aware Analysis of Catastrophic Overfitting: Revealing the Pseudo-Robust Shortcut Dependency12
Extensive experiments demonstrate that the proposed method, Layer-Aware Adversarial Weight Perturbation (LAP), can effectively prevent CO and further enhance robustness in DNNs from the CO state.
- 12
- Towards Digital Retina in Smart Cities: A Model Generation, Utilization and Communication Paradigm12
This work presents an integrated multiple deep learning models reuse and prediction strategy, which greatly increases the feasibility of the digital retina in processing and analyzing the large-scale visual data in smart cities.
- General Heterogeneous Transfer Distance Metric Learning via Knowledge Fragments Transfer12
This work develops a general and flexible heterogeneous TDML (HTDML) framework based on the knowledge fragment transfer strategy, which demonstrates superiority of the proposed method on both scene classification and object recognition.
- 11
- Alignclip: navigating the misalignments for robust vision-language generalization11
AlignCLIP is proposed, an advanced fine-tuning methodology distinguished by its attention alignment loss, designed to calibrate the distribution of attention across multi-head attention layers and introduces semantic label smoothing, a technique that leverages textual class similarities to refine prediction hierarchies.
- Can Large Language Models Help Experimental Design for Causal Discovery?11
This work presents LeGIT -- a robust framework that effectively incorporates LLMs to augment existing numerical approaches for the intervention targeting in causal discovery and demonstrates the usefulness of LLMs in assisting with experimental design for scientific discovery.
- 11
- Learning with Group Noise11
This work proposes a novel Max-Matching method for learning with group noise that utilizes a matching mechanism to evaluate the relation confidence of each object w.r.t. the target, which considers the Non-IID characteristics among objects in the group.
- 11
- Towards Mixture Proportion Estimation without Irreducibility.11
This paper builds a new irreducible MPE problem and solves the new problem, and proposes Regrouping-MPE, a method that outperforms all state-of-the-art MPE methods on various real-world datasets.
- ShadowNet for Data-Centric Quantum System Learning10
This work presents ShadowNet, implemented under both convolutional and attention mechanisms, to efficiently and faithfully tackle two pivotal QSL tasks: quantum state tomography (QST) and direct fidelity estimation (DFE).
- Finding core labels for maximizing generalization of graph neural networks10
This paper introduces and elaborate on the problem of training data selection for GNNs, and proposes the following subset hypothesis for graph data: "There exists a core subset when selecting a fixed-size dataset from the dense training dataset, and GNNs trained on this core subset can achieve a better graph representation".
- Combating Exacerbated Heterogeneity for Robust Models in Federated Learning10
A novel framework called Slack Federated Adversarial Training (SFAT), assigning the client-wise slack during aggregation to combat the intensified heterogeneity among local clients and properly relax the objective when combining federated learning and adversarial training is proposed.
- 10
- OpenInsGaussian: Open-Vocabulary Instance Gaussian Segmentation with Context-Aware Cross-View Fusion9
This paper introduces OpenInsGaussian, an Open-vocabulary Instance Gaussian segmentation framework with Context-aware Cross-view Fusion, which achieves state-of-the-art results in open-vocabulary 3D Gaussian segmentation, outperforming existing baselines by a large margin.
- 9
- Jailbreaking the Non-Transferable Barrier via Test-Time Data Disguising9
This work reveals the first loophole of black-box NTL models by proposing a novel attack method (dubbed as JailNTL) to jailbreak the non-transferable barrier through test-time data disguising.
- TED-VITON: Transformer-Empowered Diffusion Models for Virtual Try-On9
TED-VITON is proposed, a novel framework that integrates a Garment Semantic Adapter for enhancing garment-specific features, a Text Preservation Loss to ensure accurate and distortion-free text rendering, and a constraint mechanism to generate prompts by optimizing Large Language Model (LLM).
- 9
- Exploring Channel-Aware Typical Features for Out-of-Distribution Detection9
This work delve into the challenges in OOD detection from the perspective of typicality and regard the feature’s high-probability region as the feature’s typical set as the feature’s typical set.
- PNT-Edge: Towards Robust Edge Detection with Noisy Labels by Learning Pixel-level Noise Transitions9
A novel Pixel-wise Shift Learning (PSL) module is developed to estimate the transition from clean to noisy labels as a displacement field, and a local edge density regularization term is devised to exploit local structure information for better transition learning.
- Fairness Improves Learning from Noisily Labeled Long-Tailed Data9
The Fairness Regularizer is introduced, inspired by regularizing the performance gap between any two sub-populations and shows that the introduced fairness regularizer improves the performances of sub- Populations on the tail and the overall learning performance.
- Identifiability and Asymptotics in Learning Homogeneous Linear ODE Systems from Discrete Observations9
This work derives a sufficient condition for the identifiability of homogeneous linear ODE systems from a sequence of equally-spaced error-free observations sampled from a single trajectory, and proves that under mild conditions, the parameter estimator based on the NLS method is consistent and asymptotic normal with $n^{-1/2}$ convergence rate.
- 9
- 8
- 8
- Intern-GS: Vision Model Guided Sparse-View 3D Gaussian Splatting8
This work presents Intern-GS, a novel approach that effectively leverages rich prior knowledge from vision foundation models to enhance the process of sparse-view Gaussian Splatting, thereby enabling high-quality scene reconstruction.
- 8
- Optimal Kernel Choice for Score Function-based Causal Discovery8
This paper model the generative process of the variables involved in each step of the causal graph search procedure as a mixture of independent noise variables, and derives an automatic kernel selection method by maximizing the marginal likelihood of the variables involved in each search step.
- 8
- OVGaussian: Generalizable 3D Gaussian Segmentation with Open Vocabularies8
Experimental results demonstrate that OVGaussian significantly outperforms baseline methods, exhibiting robust cross-scene, cross-domain, and novel-view generalization capabilities.
- A Universal Unbiased Method for Classification from Aggregate Observations8
The goal of this paper is to present a novel universal method of CFAO, which holds an unbiased estimator of the classification risk for arbitrary losses -- previous research failed to achieve this goal.
- Robust Angular Local Descriptor Learning8
A robust angular loss is proposed which uses cosine similarity instead of L2 distance to compare descriptors and relies on a robust loss function that gives smaller penalty to triplets with negative relative similarity, which shows robustness on different datasets.
- Video Face Editing Using Temporal-Spatial-Smooth Warping8
A temporal-spatial-smooth warping method that is robust to the subtly temporal incoherence of the facial feature point localizations and is effective to preserve the temporal coherence and spatial smoothness of the control lattices for editing faces in videos is proposed.
- Strength-Adaptive Adversarial Training7
A novel technique called Strength-Adaptive Adversarial Training (SAAT), which incorporates an adversarial-loss constraint to guide the generation of adversarial training data and substantially improves adversarial robustness over standard AT.
- Defending Against Adversarial Examples Via Modeling Adversarial Noise7
This work constructs an instance-dependent label transition matrix to represent the label transition relationship for explicitly modeling adversarial noise and designs an adversarial robustness constraint at the transition matrix level to enhance the robustness of the label transition network.
- MF-VITON: High-Fidelity Mask-Free Virtual Try-On with Minimal Input7
A Mask-Free VITON (MF-VITON) framework that achieves realistic VITON using only a single person image and a target garment, eliminating the requirement for auxiliary masks, achieves state-of-the-art (SOTA) performance regarding garment transfer accuracy and visual realism.
- W-DOE: Wasserstein Distribution-Agnostic Outlier Exposure7
This work proposes a novel OE-based learning method termed Wasserstein Distribution-agnostic Outlier Exposure (W-DOE), which is both theoretically sound and experimentally superior to previous works, and suggests a general learning framework to search for the synthesized OOD data that can benefit the models most, ensuring the OOD performance for the enlarged OOD coverage measured by the Wasserstein metric.
- NVS-SQA: Exploring Self-Supervised Quality Representation Learning for Neurally Synthesized Scenes Without References7
NVS-SQA is proposed, a NSS quality assessment method to learn no-reference quality representations through self-supervision without reliance on human labels, and shows that NVS-SQA outperforms 17 no-reference methods by a large margin.
- Systematic Investigation of Sparse Perturbed Sharpness-Aware Minimization Optimizer7
This paper proposes Sparse SAM (SSAM), an efficient and effective training scheme that achieves sparse perturbation by a binary mask and theoretically proves that SSAM can converge at the same rate as SAM.
- 7
- Continual Learning From a Stream of APIs7
A novel data-free cooperative continual distillation learning framework that distills knowledge from a stream of APIs into a CL model by generating pseudo data, just by querying APIs, and proposes a new regularization term based on network similarity to prevent catastrophic forgetting of previous APIs.
- Diversity-enhancing Generative Network for Few-shot Hypothesis Adaptation7
A diversity-enhancing generative network (DEG-Net) is proposed for the FHA problem, which can generate diverse unlabeled data with the help of a kernel independence measure: the Hilbert-Schmidt independence criterion (HSIC).
- Towards Lightweight Black-Box Attacks against Deep Neural Networks7
It is argued that black-box attacks can pose practical attacks in this extremely restrictive scenario where only several test samples are available and the proposed Error TransFormer (ETF) transforms the approximation error in the parameter space into a perturbation in the feature space and alleviates the error by disturbing features.
- On the Rates of Convergence From Surrogate Risk Minimizers to the Bayes Optimal Classifier7
This article introduces the notions of consistency intensity and conductivity to characterize a surrogate loss function and exploits this notion to obtain the rate of convergence from an empirical surrogate risk minimizer to the Bayes optimal classifier, enabling fair comparisons of the excess risks of different surrogaterisk minimizers.
- 7
- 7
- Power Characterization of Noisy Quantum Kernels6
This article theoretically characterize the power of noisy quantum kernels and demonstrates that under depolarizing noise, quantum kernel methods may only have very poor prediction capability, even when the generalization error is small.
- 6
- PanoSLAM: Panoptic 3D Scene Reconstruction via Gaussian SLAM6
PanoSLAM is introduced, the first SLAM system to integrate geometric reconstruction, 3D semantic segmentation, and 3D instance segmentation within a unified framework and achieves panoptic 3D reconstruction of open-world environments directly from the RGB-D video.
- ProtoSimi: label correction for fine-grained visual categorization6
This paper proposes a novel learning framework named ProtoSimi, which employs an adaptive label correction strategy, ensuring effective learning on limited data and outperforms the existing label noisy methods by a large margin.
- Unleashing the Potential of Regularization Strategies in Learning with Noisy Labels6
It is demonstrated that a simple baseline using cross-entropy loss, combined with widely used regularization strategies like learning rate decay, model weights average, and data augmentations, can outperform state-of-the-art methods.
- 6
- A Machine Learning Approach for Predicting Human Preference for Graph Layouts6
This paper presents the first machine learning approach for predicting qualitative evaluation of graph layouts using human preference experiment data, and to the best knowledge, this is the first approach.
- 6
- 6
- 6
- GUIC: Certified Graph Unlearning with Individual Fairness Guarantees5
GUIC is introduced, the first framework that jointly ensures certified unlearning and individual fairness in graph-based models, introducing a novel perspective on responsible model updates in graph unlearning and providing a lightweight yet verifiable alternative with theoretical fairness guarantees.
- 5
- Learning Efficient and Generalizable Graph Retriever for Knowledge-Graph Question Answering5
This work proposes RAPL, a novel framework for efficient and effective graph retrieval in KGQA, and significantly reduces the performance gap between smaller and more powerful LLM-based reasoners, as well as the gap under cross-dataset settings, highlighting its superior retrieval capability and generalizability.
- DINO-Detect: A Simple yet Effective Framework for Blur-Robust AI-Generated Image Detection5
This work proposes a blur-robust AIGI detection framework based on teacher-student knowledge distillation that achieves state-of-the-art performance under both motion-blurred and clean conditions, demonstrating improved generalization and real-world applicability.
- 5
- Extracting Clean and Balanced Subset for Noisy Long-tailed Classification5
This work develops a novel pseudo labeling method using class prototypes from the perspective of distribution matching, which can be solved with optimal transport (OT) and can reduce the side-effects of noisy and long-tailed data simultaneously.
- Unraveling the Impact of Heterophilic Structures on Graph Positive-Unlabeled Learning5
A new method, named Graph PU Learning with Label Propagation Loss (GPL), which considers learning from PU nodes along with an intermediate heterophily reduction, which helps mitigate the negative impact of the heterophilic structure.
- 5
- Overcoming cohort heterogeneity for the prediction of subclinical cardiovascular disease risk5
It is demonstrated that utilizing subcohorts of individuals with subclinical CAD can improve the prediction of sub clinical CAD and can facilitate the discovery of novel biomarkers of subclinical disease.
- 5
- MSR: Making Self-supervised learning Robust to Aggressive Augmentations5
This work proposes a new SSL paradigm, which counteracts the impact of semantic shift by balancing the role of weak and aggressively augmented pairs, and can better embrace the aggressive augmentations and neutralize the semantic shift problem.
- AlphaMemo: Structured Search-Process Memory for Self-Evolving Alpha Mining Agents4
Experiments on CSI 500 and S\&P 500 show improved out-of-sample performance and fixed-budget discovery efficiency, with ablations validating the roles of residual learning, confidence gating, AST-diff motifs, and veto memory.
- Unknown-Aware Bilateral Dependency Optimization for Defending Against Model Inversion Attacks4
The bilateral dependency optimization strategy (BiDO) is proposed, a dual-objective approach that minimizes the dependency between input features and latent representations, while simultaneously maximizing the dependency between latent representations and labels.
- MOKD: Cross-domain Finetuning for Few-shot Classification via Maximizing Optimized Kernel Dependence4
A bi-level optimization framework to learn a set of class-specific representations that match the cluster structures indicated by labeled data of the given task, and can not only achieve better generalization performance on unseen domains in most cases but also learn better data representation clusters.
- 4
- 4
- Model2Scene: Learning 3D Scene Representation via Contrastive Language-CAD Models Pre-training4
This paper proposes Model2Scene, a novel paradigm that learns free 3D scene representation from Computer-Aided Design models and languages and proposes a novel feature regularization operation, termed Deep Convex-hull Regularization (DCR), to project point features into a unified convex hull space, reducing the domain gap.
- Making Binary Classification from Multiple Unlabeled Datasets Almost Free of Supervision4
This work proposes a novel framework to handle the MU-OPPO problem, which consists of four sequential modules: (i) pseudo label assignment; (ii) confident example collection; (iii) class prior estimation; (iv) classifier training with estimated class priors.
- Learning and Mining with Noisy Labels4
This tutorial summarizes the foundations and goes through the most recent noisy-label-tolerant techniques, and the audience will gain a broad knowledge of learning and mining with noisy labels from the viewpoint of statistical learning theory, deep learning, detailed analysis of typical algorithms and frameworks, and their real-world data mining applications.
- 4
- Diversified Bayesian Nonnegative Matrix Factorization4
This paper proposes to assign a diversity prior to the parts of the factorization to induce correctness based on the assumption that useful parts should be distinct and thus well-spread and aims at inducing factorizations embracing both good data fitness from maximizing likelihood and large separability from the diversity prior.
- 4
- 4
- 3
- Cross-domain Few-shot Classification via Invariant-content Feature Reconstruction3
An effective method to train a simple module that simultaneously considers both high-level and fine-grained invariant-content features for the previously unseen domains, which demonstrates the effectiveness of the fusion of the high-level features and the fine-grained invariant-content features.
- Plan before Solving: Problem-Aware Strategy Routing for Mathematical Reasoning with LLMs3
This work proposes Planning and Routing through Instance-Specific Modeling (PRISM), a novel framework that decouples mathematical reasoning into two stages: strategy planning and targeted execution and shows particularly strong benefits for mathematical reasoning tasks across diverse model architectures.
- A Knowledge-Guided Adversarial Defense for Resisting Malicious Visual Manipulation3
A knowledge-guided adversarial defense (KGAD) to actively force malicious manipulation models to output semantically confusing samples and achieves competitive performances and generalizability, indicating that it can effectively resist malicious visual manipulation.
- Provable Discriminative Hyperspherical Embedding for Out-of-Distribution Detection3
In this paper, a prototype-based contrastive learning framework for OOD detection is proposed, which is termed provable Discriminative Hyperspherical Embedding (DHE), and the maximal prototype dispersion is theoretically proved, thereby avoiding the pitfalls of local optima commonly encountered by most existing methods.
- 3
- 3
- Gradient constrained sharpness-aware prompt learning for vision-language models3
A novel SAM-based method for prompt learning is proposed, denoted as Gradient Constrained Sharpness-aware Context Optimization (GCSCoOp), to dynamically constrain the optimizing gradient, thus achieving above two-fold optimization objective simultaneously.
- Exploit CAM by itself: Complementary Learning System for Weakly Supervised Semantic Segmentation3
The Seed Reactivation to help SLM reactivate non-discriminating regions by setting a boundary for activation values, which successively identifies more regions of CAM is proposed, representing a new state-of-the-art among existing WSSS methods.
- Pluralistic Image Completion with Probabilistic Mixture-of-Experts3
A unified probabilistic graph model is introduced that represents the complex interactions in image completion and is formally established the effectiveness of the method and demonstrated it with comprehensive experiments.
- 3
- 3
- 3
- 3
- 3
- Deep Heterogeneous Multi-Task Metric Learning for Visual Recognition and Retrieval3
A deep heterogeneous MTML (DHMTML) method, in which a nonlinear mapping is learned for each task by using a deep neural network, which outperforms single-task deep metric learning algorithms and other HMTML approaches consistently on several benchmark datasets.
- 3
- Skipping Two Layers in ResNet Makes the Generalization Gap Smaller than Skipping One or No Layer3
It is shown that ResNet2 has a smaller generalization gap than ResNet1 or MLP and the applicability to the case with the ReLU activation function is confirmed.
- 2
- 2
- A Sample Efficient Conditional Independence Test in the Presence of Discretization2
This paper finds that the independence relationships of latent continuous variables can be established by addressing an over-identifying restriction problem with Generalized Method of Moments (GMM), and derives an appropriate test statistic and establishes its asymptotic distribution correctly reflecting CI by leveraging nodewise regression.
- 2
- 2
- 2
- Training-Free Robust Interactive Video Object Segmentation2
This paper proposes a training-free prompt tracking framework for interactive video object segmentation (I-PT), leveraging the powerful generalization of SAM, and introduces a cross-round space-time module (CRSTM), which adaptively aggregates mask features from previous rounds and frames, enhancing the segmentation stability.
- Advancing Counterfactual Inference through Nonlinear Quantile Regression2
This paper re-cast counterfactual reasoning as an extended quantile regression problem using neural networks and makes it possible to develop the generalization ability of the estimated counterfactual outcome to unseen data and provide an upper bound on the generalization error.
- 2
- 2
- 2
- 2
- 2
- 1
- 1
- Robust Learning from Noisily Labeled Long-Tailed Data via Fairness Regularizer1
The Fairness Regularizer is introduced, inspired by regularizing the performance gap between any two sub-populations and shows that the introduced fairness regularizer improves the performances of sub-populations on the tail and the overall learning performance.
- 1
- 1
- A Machine Learning Approach to Rapidly Downscale Sea Surface Temperature Extremes and Heat Stress on the Great Barrier Reef1
A machine learning model based on a super resolution deconvolutional neural network to rapidly downscale sea surface temperature (SST) on the Great Barrier Reef (GBR) outperforms conventional interpolation methods by capturing the spatial variability of SST and extreme thermal events.
- Improving the Instance-Dependent Transition Matrix Estimation by Exploiting Self-Supervised Learning1
- 1
- 1
- Slack Federated Adversarial Training1
This work introduces an <inline-formula>-slack mechanism to relax the original learning objective of federated adversarial training, and proposes a novel framework called SFAT (SFAT) to combat the intensified heterogeneity.
- Enhancing User-Centric Privacy Protection: An Interactive Framework through Diffusion Models and Machine Unlearning1
This study pioneers a comprehensive privacy protection framework that safeguards image data privacy concurrently during data sharing and model publication and instantiates a differential privacy diffusion model for protecting attribute information in images and a feature unlearning algorithm for efficient updates of the trained model on the revised image dataset.
- Multi-Label Noise Transition Matrix Estimation with Label Correlations: Theory and Algorithm1
This paper investigates the identifiability of class-dependent transition matrices in noisy multi-label learning and proposes a novel estimator that leverages label correlations without the need for anchor points or precise fitting of noisy class posteriors, leading to excellent classification performance.
- DeepSolo++: Let Transformer Decoder with Explicit Points Solo for Multilingual Text Spotting1
A simple DETR-like baseline that lets a single Transformer decoder with explicit points solo for multilingual text detection, recognition, and script identification all at once, and a surprisingly good extensibility of the method, in terms of character class, language type, and task is presented.
- Channel-Wise Contrastive Learning for Learning with Noisy Labels1
Channel-wise contrastive learning (CWCL) is introduced, which distinguishes authentic label information from noise by undertaking Contrastive learning across diverse channels, and tends to yield more nuanced and resilient features aligned with the authentic labels.
- Why do CNNs excel at feature extraction? A mathematical explanation1
This paper introduces a novel mathematical model for image classification, based on feature extraction, that can be used to generate images resembling real-world datasets and shows that convolutional neural network classifiers can solve discrete image classification tasks with zero error.
- 1
- 1
- 1
- Sample-Efficient Kernel Mean Estimator with Marginalized Corrupted Data1
This work proposes to corrupt data examples with noise from known distributions and presents a new kernel mean estimator, called the marginalized Kernel Mean Estimator, which estimates kernel mean under the corrupted distributions, which is sample-efficient and obtains much lower estimation errors than the existing estimators.
- 1
- MAPS: a dataset for semantic profiling and analysis of Android applications1
This paper presents the Malware Analysis and Profiling on Smartphones (MAPS) dataset, a very in-depth and detailed deep analysis of applications for extracting an enhanced set of features for analysis and training of deep neural network models.
- 1
- 1
- 1
- Diversified Dynamical Gaussian Process Latent Variable Model for Video Repair1
The diversified dynamical Gaussian process latent variable model (D2GPLVM) is proposed for considering the variety in existing videos and thus introducing a diversity encouraging prior to inducing points to ensure that the trained inducing points are more diverse and resistant for context-aware and artifacts-free based video repair.
- –
- Enhancing Practical Domain Generalization via Partially Shared Features–
A new DG framework is introduced that focuses on learning partially shared features (PSFs)-features shared among subsets of source domains, which contain and generalize ESFs, which contain and generalize ESFs.
- Privacy-Preserving Model Transcription With Differentially Private Synthetic Distillation–
This work presents privacy-preserving model transcription, a data-free model-to-model conversion solution to facilitate model deployment with a privacy guarantee, and theoretically proves that this approach can guarantee differential privacy and convergence.
- –
- Unveiling Representation Expansion: A New Lens on Out-of-Distribution Detection–
A novel perspective to address the pessimism of OOD detection by exploring representation expansion for OOD detection and formalizes a new scoring method, namely, Confidence aVerage (CoVer), which can capture the dynamic differences by simply averaging the scores obtained from different input variants.
- –
- –
- –
- On the Thinking-Language Modeling Gap in Large Language Models–
A new prompt technique termed Language-of-Thoughts (LoT) is proposed, which significantly reduces the language modeling biases in LLMs and improves the performance of LLMs across a variety of reasoning tasks.
- –
- –
- From Static to Dynamic: Adaptive Monte Carlo Search for Mathematical Process Supervision–
Adaptive Monte Carlo Search (AMCS) is a framework that transforms data generation from fixed, static to adaptive, dynamic search at the level of node value estimation and path expansion through a Monte Carlo algorithm with a temporally adaptive policy.
- –
- –
- –
- –
- –
- –
- Learning Locally, Revising Globally: Global Reviser for Federated Learning with Noisy Labels–
This paper proposes a novel method termed Federated Global Reviser (\method), a straightforward yet effective method comprising three modules that collaboratively rectify noisy labels and regularize local training that improves the label-noise robustness of FL in a self-contained manner.
- –
- –
- –
- Resultant: Incremental Effectiveness on Likelihood for Unsupervised Out-of-Distribution Detection–
The likelihood of variational DGMs is investigated and its detection performance could be improved in two directions: i) alleviating latent distribution mismatch, and ii) calibrating the dataset entropy-mutual integration.
- –
- –
- –
- –
- –
- Exploring Model Dynamics for Accumulative Poisoning Discovery–
This paper dives into the perspective of model dynamics and proposes a novel information measure, namely, Memorization Discrepancy, to explore the defense via the model-level information and proposes Discrepancies-aware Sample Correction (DSC) to defend against accumulative poisoning attacks.
- –
- –
- –
- –
- –
- –
- –
- –
- –
- –
- –
- Improving robustness of softmax corss-entropy loss via inference information–
This paper exploits inference region which inspires us to involve a margin-like inference information to SCE, resulting in a novel inference-softmax cross entropy (I-SCE) loss, which is intuitively appealing and interpretable.
- –
- –
- –
- –
- –
- The complexity of algorithmic hypothesis class–
A novel complexity measure called the algorithmic Rademacher complexity is designed to measure the algorithmmic hypothesis class of stable learning algorithms and provide sharper error bounds than the current state-of-the-art.
- –
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 2026-10-12. 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.