Label Learning Research Articles

Multi-graph embedding for partial label learning

Multi-graph embedding for partial label learning

Adaptive Collaborative Soft Label Learning for Unsupervised Multi-View Feature Selection

Unsupervised multi-view feature selection aims to select informative features with multi-view features and unsupervised learning. It is a challenging problem due to the absence of explicit semantic supervision. Recently, graph theory and hard pseudo-label learning have been adopted to solve multi-view feature selection problems under the unsupervised learning paradigm. However, graph-based methods are difficult to support large-scale real scenarios due to the high computational complexity of graph construction. Moreover, existing methods based on hard pseudo-label learning generally result in significant information loss. In this article, we propose an Adaptive Collaborative Soft Label Learning (ACSLL) model for unsupervised multi-view feature selection. In this model, collaborative soft label learning and multi-view feature selection are integrated into a unified framework. Specifically, we learn the pseudo soft labels from each view feature by a simple and efficient method and fuse them with an adaptive weighting strategy into a joint soft label matrix. This matrix is further used for guiding the feature selection process to identify valuable features. An effective optimization strategy guaranteed with proven convergence is derived to iteratively solve this problem. Experiments demonstrate the superiority of the proposed method in both feature selection accuracy and efficiency.

Attack Can Benefit: An Adversarial Approach to Recognizing Facial Expressions under Noisy Annotations

The real-world Facial Expression Recognition (FER) datasets usually exhibit complex scenarios with coupled noise annotations and imbalanced classes distribution, which undoubtedly impede the development of FER methods. To address the aforementioned issues, in this paper, we propose a novel and flexible method to spot noisy labels by leveraging adversarial attack, termed as Geometry Aware Adversarial Vulnerability Estimation (GAAVE). Different from existing state-of-the-art methods of noisy label learning (NLL), our method has no reliance on additional information and is thus easy to generalize to the large-scale real-world FER datasets. Besides, the combination of Dataset Splitting module and Subset Refactoring module mitigates the impact of class imbalance, and the Self-Annotator module facilitates the sufficient use of all training data. Extensive experiments on RAF-DB, FERPlus, AffectNet, and CIFAR-10 datasets validate the effectiveness of our method. The stabilized enhancement based on different methods demonstrates the flexibility of our proposed GAAVE.

USDNL: Uncertainty-Based Single Dropout in Noisy Label Learning

Deep Neural Networks (DNNs) possess powerful prediction capability thanks to their over-parameterization design, although the large model complexity makes it suffer from noisy supervision. Recent approaches seek to eliminate impacts from noisy labels by excluding data points with large loss values and showing promising performance. However, these approaches usually associate with significant computation overhead and lack of theoretical analysis. In this paper, we adopt a perspective to connect label noise with epistemic uncertainty. We design a simple, efficient, and theoretically provable robust algorithm named USDNL for DNNs with uncertainty-based Dropout. Specifically, we estimate the epistemic uncertainty of the network prediction after early training through single Dropout. The epistemic uncertainty is then combined with cross-entropy loss to select the clean samples during training. Finally, we theoretically show the equivalence of replacing selection loss with single cross-entropy loss. Compared to existing small-loss selection methods, USDNL features its simplicity for practical scenarios by only applying Dropout to a standard network, while still achieving high model accuracy. Extensive empirical results on both synthetic and real-world datasets show that USDNL outperforms other methods. Our code is available at https://github.com/kovelxyz/USDNL.

Can Label-Specific Features Help Partial-Label Learning?

Partial label learning (PLL) aims to learn from inexact data annotations where each training example is associated with a coarse candidate label set. Due to its practicability, many PLL algorithms have been proposed in recent literature. Most prior PLL works attempt to identify the ground-truth labels from candidate sets and the classifier is trained afterward by fitting the features of examples and their exact ground-truth labels. From a different perspective, we propose to enrich the feature space and raise the question ``Can label-specific features help PLL?'' rather than learning from examples with identical features for all classes. Despite its benefits, previous label-specific feature approaches rely on ground-truth labels to split positive and negative examples of each class and then conduct clustering analysis, which is not directly applicable in PLL. To remedy this problem, we propose an uncertainty-aware confidence region to accommodate false positive labels. We first employ graph-based label enhancement to yield smooth pseudo-labels and facilitate the confidence region split. After acquiring label-specific features, a family of binary classifiers is induced. Extensive experiments on both synthesized and real-world datasets are conducted and the results show that our method consistently outperforms eight baselines. Our code is released at https://github.com/meteoseeker/UCL

Polyphonic piano transcription based on graph convolutional network

The automatic music transcription (AMT) task is designed to convert raw performance audio signals into digital representations of symbolic music for possible computational musicology. In polyphonic music, there are multiple notes and they may appear at the same time. The combined output of multiple notes faces the problem of dimension explosion, which makes it difficult to achieve accurate transcription. To overcome the above challenge, a deep learning model based on graph convolution network (CR-GCN) is proposed to cope with the problem of dimension explosion by exploring the interdependence between musical notes. The model is divided into two parts, i.e., feature learning and label learning. Feature learning is composed of serial convolutional neural network (CNN) and recurrent neural network (RNN), which aims to extract temporal and spatial features from the input music signals. Label learning consists of graph convolutional network (GCN), which is adopted to model the interdependence between notes. The CR-GCN model can be end-to-end trainable through joint training feature networks and label networks. Experiments on public polyphonic music data sets show that the proposed method is able to mine more co-existing notes, and it is superior to existing methods in both frame-level and note-level indexes. Moreover, visual analysis shows that the learned interdependence between notes has good explainability in music theory.

Context Label Learning: Improving Background Class Representations in Semantic Segmentation.

Background samples provide key contextual information for segmenting regions of interest (ROIs). However, they always cover a diverse set of structures, causing difficulties for the segmentation model to learn good decision boundaries with high sensitivity and precision. The issue concerns the highly heterogeneous nature of the background class, resulting in multi-modal distributions. Empirically, we find that neural networks trained with heterogeneous background struggle to map the corresponding contextual samples to compact clusters in feature space. As a result, the distribution over background logit activations may shift across the decision boundary, leading to systematic over-segmentation across different datasets and tasks. In this study, we propose context label learning (CoLab) to improve the context representations by decomposing the background class into several subclasses. Specifically, we train an auxiliary network as a task generator, along with the primary segmentation model, to automatically generate context labels that positively affect the ROI segmentation accuracy. Extensive experiments are conducted on several challenging segmentation tasks and datasets. The results demonstrate that CoLab can guide the segmentation model to map the logits of background samples away from the decision boundary, resulting in significantly improved segmentation accuracy. Code is available.

Combining layered label correction and mixup supervised contrastive learning to learn noisy labels

Deep neural networks trained using standard cross-entropy loss often induce overfitting of noisy labels that degrades performance. Label correction and contrastive learning have been theoretically demonstrated to be mutually beneficial in noisy label learning. However, the fusion of these two modules in practice has not been researched adequately, and the influence of each module on performance improvement has not been clarified. In this paper, we propose an effective noisy label learning framework that combines layered label correction with mixup supervised contrastive learning, based on hierarchical clustering using first neighbor relations and mixup contrastive loss. In particular, the framework consists of two modules—Layered Label Correction (LLC) and Mixup Supervised Contrastive Loss (MSCL). The LLC module aggregates the first neighbor's soft labels via label propagation among multilayer partitions based on the hierarchical clustering of the feature representation. Its purpose is to capture the underlying structure of learning representations at different levels of granularity. The MSCL module incorporates pseudo-labels into mixup contrastive learning directly to learn class-related representations of all instances of each class. Ablation studies are conducted to evaluate the contribution of individual modules to the overall performance enhancement of the framework. Moreover, experiments on synthetic datasets comprising noise of various patterns and levels demonstrate that the proposed framework yields a superior classification accuracy in comparison with existing alternatives.

Cross-modal hashing with missing labels

Hashing-based cross-modal retrieval methods have become increasingly popular due to their advantages in storage and speed. While current methods have demonstrated impressive results, there are still several issues that have not been addressed. Specifically, many of these approaches assume that labels are perfectly assigned, despite the fact that in real-world scenarios, labels are often incomplete or partially missing. There are two reasons for this, as manual labeling can be a complex and time-consuming task, and annotators may only be interested in certain objects. As such, cross-modal retrieval with missing labels is a significant challenge that requires further attention. Moreover, the similarity between labels is frequently ignored, which is important for exploring the high-level semantics of labels. To address these limitations, we propose a novel method called Cross-Modal Hashing with Missing Labels (CMHML). Our method consists of several key components. First, we introduce Reliable Label Learning to preserve reliable information from the observed labels. Next, to infer the uncertain part of the predicted labels, we decompose the predicted labels into latent representations of labels and samples. The representation of samples is extracted from different modalities, which assists in inferring missing labels. We also propose Label Correlation Preservation to enhance the similarity between latent representations of labels. Hash codes are then learned from the representation of samples through Global Approximation Learning. We also construct a similarity matrix according to predicted labels and embed it into hash codes learning to explore the value of labels. Finally, we train linear classifiers to map original samples to a low-dimensional Hamming space. To evaluate the efficacy of CMHML, we conduct extensive experiments on four publicly available datasets. Our method is compared to other state-of-the-art methods, and the results demonstrate that our model performs competitively even when most labels are missing.

Simple and Effective Complementary Label Learning Based on Mean Square Error Loss

In this paper, we propose a simple and effective complementary label learning approach to address the label noise problem for deep learning model. Different surrogate losses have been proposed for complementary label learning, however, are often sophisticated designed, as the losses are required to satisfy the classifier consistency property. We propose an effective square loss for complementary label learning under unbiased and biased assumptions. We also show theoretically that our method assurances that the optimal classifier under complementary labels is also the optimal classifier under ordinary labels. Finally, we test our method on three different benchmark datasets with biased and unbiased assumptions to verify the effectiveness of our method.

Multi-view prototype-based disambiguation for partial label learning

In this work, we study the multi-view partial label learning (MVPLL) problem, where each instance is depicted by different view features and associated with a set of candidate labels, among which a true label exists but is inaccessible in the training phase. Most existing PLL methods only consider single-view case, which learn view classifier independently and neglect the view correlations, thus can not be applied to solve MVPLL problem. Due to the non-deep framework, traditional MVPLL approach is weak in the representation ability, so its performance is still to be improved. To solve the MVPLL problem, a deep multi-view prototype-based disambiguation approach is proposed in this paper. Specifically, we innovatively employ the deep neural network for multi-view ambiguously-labeled image classification to enhance the representation ability, which makes use of the information fusion between multiple views. To improve the discriminative ability, we propose multi-view prototype-based label disambiguation algorithm. On theoretical aspect, an estimation error bound for view-risk estimator is established, which is shown to be larger than that for fuse-risk estimator. Experiments demonstrate the superiorities of our proposed method in terms of the prediction accuracy.

SPL-LDP: a label distribution propagation method for semi-supervised partial label learning

SPL-LDP: a label distribution propagation method for semi-supervised partial label learning

English-Learning 12-Month-Olds Do Not Map Function-Like Words to Objects

ABSTRACT By 12 months, English-learning infants have an awareness of the sound patterns of word forms that constitute acceptable labels for objects in their native language. In the following experiments, we replicated and extended previous findings that Canadian English-learning infants will not link function-like words with novel objects. Across three experiments using the Switch task, 101 infants living in Calgary, Canada, were habituated to two CV and VC word-object pairings. At test, infants did not look longer on the Switch trial and the Same trial, suggesting they did not form word-object associations between prototypical function words and a novel object in any of the experiments (ps>0.5). This set of null results extends prior research showing that Canadian English-learning infants will not link function-like words with novel objects and suggests that infants’ prior experience with their native language may constrain their learning of novel labels.

Unsupervised Radio Frequency Fingerprint Identification Based on Curriculum Learning

Radio frequency fingerprint (RFF) identification aims to identify emitters by extracting the inherent physical-layer features. Despite the significant accuracy achieved in a supervised learning manner, few works have focused on unsupervised RFF identification. To this end, we propose an unsupervised training paradigm based on the curriculum learning framework. Specifically, the unlabeled samples are first sorted by the signal-to-noise (SNR) of signals. Then, the network is pretrained with contrastive learning on the high SNR subset. To further improve identification performance, the model is gradually optimized from the easy (high SNR) to the hard (low SNR) subsets with pseudo label learning. Experimental results indicate that the proposed scheme achieves comparable performance against schemes based on supervised learning.

Inaccurate-Supervised Learning With Generative Adversarial Nets.

Inaccurate-supervised learning (ISL) is a weakly supervised learning framework for imprecise annotation, which is derived from some specific popular learning frameworks, mainly including partial label learning (PLL), partial multilabel learning (PML), and multiview PML (MVPML). While PLL, PML, and MVPML are each solved as independent models through different methods and no general framework can currently be applied to these frameworks, most existing methods for solving them were designed based on traditional machine-learning techniques, such as logistic regression, KNN, SVM, decision tree. Prior to this study, there was no single general framework that used adversarial networks to solve ISL problems. To narrow this gap, this study proposed an adversarial network structure to solve ISL problems, called ISL with generative adversarial nets (ISL-GANs). In ISL-GAN, fake samples, which are quite similar to real samples, gradually promote the Discriminator to disambiguate the noise labels of real samples. We also provide theoretical analyses for ISL-GAN in effectively handling ISL data. In this article, we propose a general framework to solve PLL, PML, and MVPML, while in the published conference version, we adopt the specific framework, which is a special case of the general one, to solve the PLL problem. Finally, the effectiveness is demonstrated through extensive experiments on various imprecise annotation learning tasks, including PLL, PML, and MVPML.

Hierarchical multimodal fusion framework based on noisy label learning and attention mechanism for cancer classification with pathology and genomic features.

Classification of subtype and grade is imperative in the clinical diagnosis and prognosis of cancer. Many deep learning-based studies related to cancer classification are based on pathology and genomics. However, most of them are late fusion-based and require full supervision in pathology image analysis. To address these problems, we present an integrated framework for cancer classification with pathology and genomics data. This framework consists of two major parts, a weakly supervised model for extracting patch features from whole slide images (WSIs), and a hierarchical multimodal fusion model. The weakly supervised model can make full use of WSI labels, and mitigate the effects of label noises by the self-training strategy. The generic multimodal fusion model is capable of capturing deep interaction information through multi-level attention mechanisms and controlling the expressiveness of each modal representation. We validate our approach on glioma and lung cancer datasets from The Cancer Genome Atlas (TCGA). The results demonstrate that the proposed method achieves superior performance compared to state-of-the-art methods, with the competitive AUC of 0.872 and 0.977 on these two datasets respectively. This paper establishes insight on how to build deep networks on multimodal biomedical data and proposes a more general framework for pathology image analysis without pixel-level annotation.

Caregiver speech predicts the emergence of children's emotion vocabulary.

Learning about emotions is an important part of children's social and communicative development. How does children's emotion-related vocabulary emerge over development? How may emotion-related information in caregiver input support learning of emotion labels and other emotion-related words? This investigation examined language production and input among English-speaking toddlers (16-30 months) using two datasets: Wordbank (N=5520; 36% female, 38% male, and 26% unknown gender; 1% Asian, 4% Black, 2% Hispanic, 40% White, 2% others, and 50% unknown ethnicity; collected in North America; dates of data collection unknown) and Child Language Data Exchange System (N=587; 46% female, 44% male, 9% unknown gender, all unknown ethnicity; collected in North America and the UK; data collection dates, were available between 1962 and 2009). First, we show that toddlers develop the vocabulary to express increasingly wide ranges of emotional information during the first 2 years of life. Computational measures of word valence showed that emotion labels are embedded in a rich network of words with related valence. Second, we show that caregivers leverage these semantic connections in ways that may scaffold children's learning of emotion and mental state labels. This research suggests that young children use the dynamics of language input to construct emotion word meanings, and provides new techniques for defining the quality of infant-directed speech.

Prior Knowledge Constrained Adaptive Graph Framework for Partial Label Learning

Partial label learning (PLL) aims to learn a robust multi-class classifier from the ambiguous data, where each instance is given with several candidate labels, among which only one label is real. Most existing methods usually cope with such problem by utilizing a feature similarity graph to conduct label disambiguation. However, these methods construct the feature graph by only employing original features, while the influences of latent outliers and the contributions of label space are regrettably ignored. To tackle these issues, in this article, we propose a P rior Kn O wledge Cons T rained A daptive G raph Fram E work ( POTAGE ) for partial label learning, which utilizes an adaptive graph fused with label information to accurately describe the instance relationship and guide the desired model training. Compared with the feature-induced fixed graph, the adaptive graph is deemed to be more robust and accurate to reveal the intrinsic manifold structure within the data, and the embedding label information is expected to effectively alleviate the label ambiguities and enlarge the gap of label confidences between two instances from different classes. Extensive experiments demonstrate that POTAGE achieves state-of-the-art performance.

Semi-Supervised Multi-Label Dimensionality Reduction Learning by Instance and Label Correlations

The label learning mechanism is challenging to integrate into the training model of the multi-label feature space dimensionality reduction problem, making the current multi-label dimensionality reduction methods primarily supervision modes. Many methods only focus attention on label correlations and ignore the instance interrelations between the original feature space and low dimensional space. Additionally, very few techniques consider how to constrain the projection matrix to identify specific and common features in the feature space. In this paper, we propose a new approach of semi-supervised multi-label dimensionality reduction learning by instance and label correlations (SMDR-IC, in short). Firstly, we reformulate MDDM which incorporates label correlations as a least-squares problem so that the label propagation mechanism can be effectively embedded into the model. Secondly, we investigate instance correlations using the k-nearest neighbor technique, and then present the l1-norm and l2,1-norm regularization terms to identify the specific and common features of the feature space. Experiments on the massive public multi-label data sets show that SMDR-IC has better performance than other related multi-label dimensionality reduction methods.

Infants use contextual memory to attend and learn in naturalistic scenes.

Infants encounter new objects and learn about object features in relation to a rich and detailed visuospatial context. Using a contextual cueing task, recent work showed that 6- and 10-month-old infants search more efficiently for target objects in repeated rather than novel visuospatial contexts (i.e., arrays of shapes on a blank background). Here, we investigate whether infants' sensitivity to visuospatial context scales up to more complex and potentially more distracting, naturalistic scenes. In an eye-tracking task, 8-month-olds searched for a novel target object in colorful photographs of everyday environments (e.g., bedrooms and kitchens). Repeated ("Old") contexts co-varied with target locations, such that the target object appeared in exactly the same location on the same scene, while varying ("New") contexts contained target objects placed in different counterbalanced locations across a variety of scenes. Infants exhibited faster search times, more anticipation of target animation, and longer looking at targets that appeared in Old relative to New contexts. In a subsequent memory test, infants showed better recognition of label-object pairings for target objects that had appeared in Old, rather than New, contexts. These results indicate that infants can use visuospatial contextual information in complex naturalistic scenes to facilitate memory-guided attention and learning of object-paired labels.