Deep Neural Networks (DNNs) have achieved great success on classification tasks of closed class sets. However, new classes, like new categories of social media topics, are continual added to the real world, making it necessary to learn ...

• A two-stage framework for class incremental learning (CIL) with a pre-trained feature encoder yielding high intrinsic dimensionality is proposed.
• An Incremental Prototype Classifier (IPC) is used to address classifier distortion.
• ...

Deep neural networks (DNNs) are easily biased towards the training set, which causes substantial performance degradation for out-of-distribution data. Many methods are studied to generalize under various distribution shifts in the literature of ...

Display Omitted

• A large-scale Chinese character recognition dataset for a practical printed-to- handwritten recognition task is proposed, which reflects distribution shifts in real- world applications.
• Extensive experiments on the proposed PaHCC (...

The impressive success of style-based GANs (StyleGANs) in high-fidelity image synthesis has motivated research to understand the semantic properties of their latent spaces. In this paper, we approach this problem through a geometric analysis of ...

• We propose a scheme to estimate local intrinsic dimensions in GAN latent spaces.
• Our local dimension estimate provides an upper bound on local semantic perturbations.
• We propose a layer-wise unsupervised disentanglement score, ...

Traditional signal processing methods relying on mathematical data generation models have been cast aside in favour of deep neural networks, which require vast amounts of data. Since the theoretical sample complexity is nearly impossible to ...

• Neural networks systematically outperform traditional expert methods when trained on sufficiently large datasets, yet the necessary amount of training data has been largely overlooked.
• We study three well-posed learning problems ...

In this paper, we propose a novel method to address the challenge of learning deep neural network models in the presence of open-set noisy labels, which include mislabeled samples from out-of-distribution categories. Previous methods relied on ...

• Propose a multi-prototype modeling mechanism for learning with noisy labels.
• Use cross-augmentation and siamese losses for training.
• Outperform others in tests on CIFAR100, Clothing1M and Food101N.

Dynamic ensemble selection (DES) effectively improves aggregation performance by dynamically finding the most appropriate subset of classifiers for each query sample. The key component of this strategy is to define an appropriate criterion for ...

• A new generalized group-based selection criterion, synergy competence, is introduced.
• An algorithm Shapley is proposed to assess candidate classifiers’ synergy competence.
• A new dynamic ensemble selection method is developed based ...

Vision Transformer (ViT) has achieved unprecedented success in vision tasks with the assistance of abundant data. However, the lack of inductive bias in lightweight ViT makes learning locality challenging on small datasets, leading to poor ...

• The Lightweight Vision Transformer with knowledge distillation can excel on small datasets.
• Knowledge distillation combined with Curriculum Learning can enhance distillation efficiency.
• Feature-based knowledge distillation can ...

Ensemble learning is one of the best solutions for imbalanced classification problems. Diversity is a key factor that affects the performance of ensemble learning. Most existing diversity metrics such as Q-statistics measure diversity based on ...

• A data-level diversity measure IED to evaluate diversity without model training.
• A new imbalanced ensemble learning model based on data-level diversity IED.
• A new weight-adaptive voting strategy for ensemble learning.
• Strong ...

The skewed class distribution and data complexity may severely affect the imbalanced classification results. The cost of classification can be significantly reduced if these data complexity are measured and pre-processed prior to training, ...

• MFII considers class imbalance and various overlap factors to assess data complexity.
• VoR and DoC are proposed to estimate resolution and stability of complexity measures.
• MFII has strong negative correlation with classification ...

In current classification tasks, an important method to improve accuracy is to pre-train the model using a large-scale domain-specific dataset. However, many tasks such as writer identification (writerID) lack suitable large-scale datasets in ...

• The proposed method utilizes information from top-k outputs to boost the accuracy of top-1 results without vast pre-training process.
• This efficient post-processing framework can rectify the probability of the correct class being ...

This paper proposes a probabilistic deep metric learning (PDML) framework for hyperspectral image classification (HSIC). The core problem for HSIC is the spectral variability between intraclass materials and the spectral similarity between ...

• We adopt probabilistic embedding to consider both spectral and label uncertainty.
• We generate samples by Monte Carlo sampling and impose a DML loss for optimization.
• We conduct experiments on 4 widely used datasets and achieve the ...

Efficiency and trustworthiness are two eternal pursuits when applying deep learning in practical scenarios. Considering efficiency, dataset distillation (DD) endeavors to reduce training costs by distilling large datasets into tiny ones. However, ...

• We propose a novel paradigm called TrustDD, ensuring both efficiency and trustworthiness.
• The proposed POE surpasses SOTA OE even without real outlier data for OOD detection.
• Experiments show TrustDD improves OOD detection without ...

With the surge in realistic text tampering, detecting fraudulent text in images has gained prominence for maintaining information security. However, the high costs associated with professional text manipulation and annotation limit the ...

• A new dataset for text manipulation detection with diverse handcraft manipulations
• An asymmetric dual-stream baseline framework to exploit different transformed domains
• An aggregation hub and a fusion module for efficient multi-...

Compositional Zero-Shot Learning (CZSL) aims to recognize seen and unseen attribute-object compositions. Recently, some researchers apply vision-language models to CZSL task. However, they only roughly match the image embedding and composition ...

• We present a novel perspective that a visual primitive can be regarded as a word.
• We propose VisPrompt for interacting visual primitives with sub-concepts in a prompt.
• VisPrompt can easily align visual elements with text and ...

Progressive training has unfolded its superiority on a wide range of downstream tasks. However, it may fail in fine-grained recognition (FGR) due to special challenges with high intra-class and low inter-class variances. In this paper, we propose ...

• We are the first to exploit the sub–modularity for active sample selection. By our problem formulation, the optimal category subsets can be progressively selected for obtaining steady cumulative gain.
• We combine submodular optimization ...

In sensor-based Human Activity Recognition (HAR), a predominant assumption is that the data utilized for training and evaluation purposes are drawn from the same distribution. It is also assumed that all data samples are independent and ...

• A novel DGDATA framework is proposed and designed for sensor-based cross-user HAR.
• The framework uses adversarial learning with temporal attention and generative model.
• A CVAE generative model is used to align data distribution ...

Few-Shot Learning (FSL) is a challenging classification task in machine learning, and it aims to recognize unseen examples of new classes with only a few labeled reference examples (i.e., the support set). The training phase of FSL typically ...

• We propose a new problem setting termed FSL-LTL to consider a frequently occurring practical issue in which the class labels are long-tailed.
• We build a novel two-stage training framework called RCE to solve this new problem. It ...

• The FA-adapter is presented to construct feature similarity densely for more reliable results of category predictions.
• To promote the training efficiency for our proposed FA-adapter, a two-stage training phase consisting of backbone ...

Few-shot learning aims to transfer the knowledge learned from seen categories to unseen categories with a few references. It is also an essential challenge to bridge the gap between humans and deep learning models in real-world applications. ...

As an inevitable issue in annotating large-scale datasets, instance-dependent label noise (IDN) can cause serious overfitting in neural networks. To combat IDN, label reconstruction methods have been developed with noise transition matrices or ...

• We propose a dynamic sample selection method when performing label reconstruction.
• The proposed method is adaptive to different noise rates.
• Instance features are used along with classifier outputs to reconstruct noisy labels.

Codebook collapse is a common problem in training deep generative models with discrete representation spaces like Vector Quantized Variational Autoencoders (VQ-VAEs). We observe that the same problem arises for the alternatively designed discrete ...

• We report evidence of the confirmation bias problem caused by the softmax in dVAE.
• We demonstrate that the confirmation bias causes the codebook collapse in dVAE.
• We propose an hierarchical extension of dVAE with an evidential view ...