Partial Label Learning Research Articles

An Unbiased Risk Estimator for Partial Label Learning with Augmented Classes

Partial Label Learning (PLL) is a typical weakly supervised learning task, which assumes each training instance is annotated with a set of candidate labels containing the ground-truth label. Recent PLL methods adopt identification-based disambiguation to alleviate the influence of false positive labels and achieve promising performance. However, they require all classes in the test set to have appeared in the training set, ignoring the fact that new classes will keep emerging in real applications. To address this issue, in this paper, we focus on the problem of Partial Label Learning with Augmented Class (PLLAC), where one or more augmented classes are not visible in the training stage but appear in the inference stage. Specifically, we propose an unbiased risk estimator with theoretical guarantees for PLLAC, which estimates the distribution of augmented classes by differentiating the distribution of known classes from unlabeled data and can be equipped with arbitrary PLL loss functions. Besides, we provide a theoretical analysis of the estimation error bound of the estimator, which guarantees the convergence of the empirical risk minimizer to the true risk minimizer as the number of training data tends to infinity. Furthermore, we add a risk-penalty regularization term in the optimization objective to alleviate the influence of the over-fitting issue caused by negative empirical risk. Extensive experiments on benchmark, UCI and real-world datasets demonstrate the effectiveness of the proposed approach.

Dealing with partial labels by knowledge distillation

Partial label learning (PLL) is a weakly supervised methodology dealing with tasks that have annotation problems by replacing the single label with a collection of candidate labels. Compared to single labels, utilizing partial labels faces challenges: (1) The limited supervision and sensitivity to the false positive candidates; (2) Situations where the ground truth is not in the candidate label sets (noisy PLs). However, in the case that there exists a subset of samples that can be easily labeled (referred to as clean samples), the existing PLL paradigm needlessly assigns these instances with partial labels randomly. To better utilize the clean samples, and alleviate the obstacles of adopting partial labels, we proposed a specific Partial Label Knowledge Distillation (PLKD) framework to distill the knowledge from the samples with low annotating cost, further guiding partial label samples with limited supervision in these scenarios. The teacher model of PLKD was pre-trained on the clean samples with a single label, which can reduce the effect of noisy PLs when training on the remaining PLL samples. Additionally, recognizing that the existing candidate labels are sampled under the uniform distribution, which may not reflect real-life scenarios, we also proposed a label-specific candidate generation method. Correspondingly, a new loss function based on our generation method is presented to evaluate the distinction between partial labels and predictions. Furthermore, we also present a partial-label guided version, denoted as PLKD-pl, to alleviate the teacher’s risk of over-confidence when the distribution between the clean set and partial label set varies widely. Extensive experimental evaluations have been conducted to demonstrate the superiority of PLKD over six state-of-the-art counterparts.

Partial label learning via weighted centroid clustering disambiguation

Partial Label Learning (PLL) is a weakly supervised learning problem that induces a multi-class classifier from data with candidate labels, among which only one is the ground-truth label. The crucial challenge in PLL is to disambiguate the false-positive labels from the candidate labels. However, most existing PLL methods fail to simultaneously consider both the instance-level similarity and the class-level information during label disambiguation. In this paper, we propose a novel two-stage method based on weighted centroid clustering, which efficiently utilizes both the local similarity among instances and the per-class global information. In the first stage, we initialize the center of each class using label propagation based on the instance-level similarity, and obtain the disambiguated instances via weighted centroid clustering derived from the per-class global information. In the second stage, the disambiguated instances are used to train a multi-class classifier. Extensive experiments on both controlled UCI datasets and real-world datasets show the superiority of the proposed method in classification accuracy.

Partial label learning via identifying outlier features

Partial label learning (PLL) is a classification problem where each training instance is ambiguously annotated with a set of candidate labels, among which only one is the ground truth label. The topological structure in the feature space is frequently used by existing PLL approaches to clarify if a candidate label is the ground truth label for a training example. However, these techniques frequently experience cumulative errors brought on by the error-prone labeling confidence estimate throughout the topological structure because of the redundant and noisy features occurring naturally in the feature space. In this paper, we propose a novel approach through Identifying Outlier Features (PL-IOF) to address this challenge. The feature space is decomposed into class prototype features and outlier features, where the former captures the discriminative features of each class prototype, and the latter captures the outlier features in each training instance. A unified framework is specifically suggested to simultaneously optimize class prototypes and outlier features, as well as to estimate labeling confidences over incomplete label training instances. This framework ensures the high quality of the extracted class prototypes by recognizing the outliers. Experiments on both synthetic and real-world datasets demonstrate the out-performance of PL-IOF over the state-of-the-art.

Partial Label Learning with Noisy Labels

Partial Label Learning with Noisy Labels

Multi-kernel partial label learning using graph contrast disambiguation

Multi-kernel partial label learning using graph contrast disambiguation

Partial label feature selection via label disambiguation and neighborhood mutual information

Partial label learning aims to learn from training instances, each of which is associated with a set of candidate labels but only one is a ground-truth label. Feature selection is an effective method to improve the generalization capability of the learning model; however, partial label feature selection work is exceptionally challenging due to the limitation and ambiguity of label information. Therefore, this paper proposes a partial label feature selection algorithm based on label disambiguation and neighborhood mutual information. Firstly, neighborhood granularity is utilized to determine the neighborhoods of instances to disambiguate the candidate labels. Secondly, based on label confidence induced by disambiguation, feature relevance and redundancy are measured by neighborhood mutual information, which avoids the negative impact of data discretization on feature selection and directly handles continuous features. Concurrently, the kappa coefficient is employed to estimate the label consistency for describing the influences of feature changes on the label space. Then, the significance of each feature is evaluated by fusing feature relevance, feature redundancy, and label consistency. Finally, the effectiveness of the proposed algorithm is verified by comparing the proposed algorithm with four base classifiers and other feature selection methods. Furthermore, the feasibility of the proposed disambiguation method is demonstrated through comparison with four state-of-the-art disambiguation methods.

Self-paced method for transfer partial label learning

In partial label learning (PLL) problem, each training sample corresponds to a group of candidate labels, in which only one label is the ground-truth label (correct label). Almost all the existing PLL algorithms attempt to eliminate the ambiguity of the candidate label sets by treating all labels indiscriminately. However, this kind of approach lacks the consideration of the complexity between the labels and the instances in the training process. Encouraged by the extensive researches of the self-paced learning (SPL) and transfer learning (TL) in various fields, this paper introduces SPL and TL together to address the PLL problem and proposes a new SPL framework, which is called self-paced method for transfer partial label learning (SPTPLL). The proposed model utilizes transfer learning model to share the parameters and regularization terms of the Support Vector Machine (SVM), which can transfer knowledge from the source task to the target task. Additionally, we implement the self-paced learning scheme by choosing a suitable self-paced function to enhance the robustness of the proposed model. In the process of learning iteration, the priority of training examples with their candidate labels is ranked through self-paced learning to control the learning process. Finally, we demonstrate the superior performance of the proposed method through a large number of experiments compared with state-of-the-art baseline methods.

Partial label learning with heterogeneous domain adaptation

Partial label learning (PLL) seeks a classification model with partially labeled (PL) instances. Each PL instance is attached with a candidate label set, with only one being ground-truth. A fundamental assumption of the previous PLL works is that there are sufficient PL instances in the learning process. Nevertheless, this assumption may not always be valid. In real-world scenarios, there may be only a few PL instances available for training. To this end, we introduce a new PLL method with heterogeneous domain adaptation (PLL-HDA). PLL-HDA leverages the knowledge from the source domain to induce a PLL classifier of the target domain, which contains only a few PL instances for training. Firstly, PLL-HDA introduces a common subspace to measure the instances from the source and target domains, where the instances from these domains are described by heterogeneous features of distinct dimensions. Secondly, we augment the features of the common subspace by merging the original features from both domains. Thirdly, a large-margin-based PLL learning system is established on the augmented features. It leads to a simplified dual form, which can be directly resolved by off-the-shelf support vector machine solvers. Lastly, a heuristic framework is proposed to resolve the PLL-HDA classification model. In this framework, the tasks of learning the classifier on the augmented features and identifying the ground-truth labels of target domain are conducted alternately. Extensive experiments on both the controlled PLL and real-world PLL datasets illustrate the superiority of PLL-HDA over the existing PLL methods. Our code is available at: https://github.com/Sux86.

On the Robustness of Average Losses for Partial-Label Learning.

Partial-label learning (PLL) utilizes instances with PLs, where a PL includes several candidate labels but only one is the true label (TL). In PLL, identification-based strategy (IBS) purifies each PL on the fly to select the (most likely) TL for training; average-based strategy (ABS) treats all candidate labels equally for training and let trained models be able to predict TL. Although PLL research has focused on IBS for better performance, ABS is also worthy of study since modern IBS behaves like ABS in the beginning of training to prepare for PL purification and TL selection. In this paper, we analyze why ABS was unsatisfactory and propose how to improve it. Theoretically, we propose two problem settings of PLL and prove that average PL losses (APLLs) with bounded multi-class losses are always robust, while APLLs with unbounded losses may be non-robust, which is the first robustness analysis for PLL. Experimentally, we have two promising findings: ABS using bounded losses can match/exceed state-of-the-art performance of IBS using unbounded losses; after using robust APLLs to warm start, IBS can further improve upon itself. Our work draws attention to ABS research, which can in turn boost IBS and push forward the whole PLL.

PiCO+: Contrastive Label Disambiguation for Robust Partial Label Learning.

Partial label learning (PLL) is an important problem that allows each training example to be labeled with a coarse candidate set with the ground-truth label included. However, in a more practical but challenging scenario, the annotator may miss the ground-truth and provide a wrong candidate set, which is known as the noisy PLL problem. To remedy this problem, we propose the PiCO+ framework that simultaneously disambiguates the candidate sets and mitigates label noise. Core to PiCO+, we develop a novel label disambiguation algorithm PiCO that consists of a contrastive learning module along with a novel class prototype-based disambiguation method. Theoretically, we show that these two components are mutually beneficial, and can be rigorously justified from an expectation-maximization (EM) algorithm perspective. To handle label noise, we extend PiCO to PiCO+, which further performs distance-based clean sample selection, and learns robust classifiers by a semi-supervised contrastive learning algorithm. Beyond this, we further investigate the robustness of PiCO+ in the context of out-of-distribution noise and incorporate a novel energy-based rejection method for improved robustness. Extensive experiments demonstrate that our proposed methods significantly outperform the current state-of-the-art approaches in standard and noisy PLL tasks and even achieve comparable results to fully supervised learning.

Label disambiguation-based feature selection for partial label learning via fuzzy dependency and feature discernibility

Partial label learning is a multi-class classification issue in which each training instance is associated with a set of candidate labels. Feature selection is an effective method to improve the performance of the learning model, and at the same time, feature selection being a challenging problem in partial label learning due to the label ambiguity of partial labels. To tackle this challenge, this paper proposes a label disambiguation-based feature selection for partial label learning via fuzzy dependency and feature discernibility. Specifically, considering the high sensitivity of the fuzzy rough sets model to pseudo labels, a instance distribution-based label disambiguation method is presented to reduce the noise from the candidate labels. Based on this, a weighted fuzzy rough sets model is constructed in accordance with the distribution information of labels, and the function of fuzzy dependency is redefined. Then, the evaluation function of feature significance is obtained by fusing fuzzy dependency and feature discernibility for identifying the critical features. Finally, extensive experiments have confirmed the feasibility and effectiveness of the proposed method. Compared to other feature selection algorithms, the proposed method exhibits superior performance and enhances the generalization of partial label learning models. Furthermore, the feasibility of the proposed label disambiguation method is demonstrated through comparison with state-of-the-art label disambiguation methods.

Blood Cell Attribute Classification Algorithm Based on Partial Label Learning

Hematological morphology examinations, essential for diagnosing blood disorders, increasingly utilize deep learning. Blood cell classification, determined by combinations of cell attributes, is complicated by the complex relationships and subtle differences among the attributes, resulting in significant time and cost penalties. This study introduces the Partial Label Learning for Blood Cell Classification (P4BC) strategy, a method that trains neural networks using the blood cell attribute labeling data of weak annotations. Using morphological knowledge, we predefined candidate label sets for the blood cell attributes to blend this knowledge with deep learning. This improves the model’s prediction accuracy and interpretability in classifying attributes. This method effectively combines morphological knowledge with deep learning, an approach we refer to as knowledge alignment. It results in an 8.66% increase in attribute recognition accuracy and a 1.09% improvement in matching predictions to the candidate label sets, compared to the original method. These results confirm our method’s ability to grasp the characteristic information of blood cell attributes, enhancing the model interpretability and achieving knowledge alignment between hematological morphology and deep learning. Our algorithm ensures attribute classification accuracy and shows excellent cell category classification, highlighting its wide application potential and practical value in blood cell category classification.

Partial label learning for automated classification of single-cell transcriptomic profiles.

Single-cell RNA sequencing (scRNASeq) data plays a major role in advancing our understanding of developmental biology. An important current question is how to classify transcriptomic profiles obtained from scRNASeq experiments into the various cell types and identify the lineage relationship for individual cells. Because of the fast accumulation of datasets and the high dimensionality of the data, it has become challenging to explore and annotate single-cell transcriptomic profiles by hand. To overcome this challenge, automated classification methods are needed. Classical approaches rely on supervised training datasets. However, due to the difficulty of obtaining data annotated at single-cell resolution, we propose instead to take advantage of partial annotations. The partial label learning framework assumes that we can obtain a set of candidate labels containing the correct one for each data point, a simpler setting than requiring a fully supervised training dataset. We study and extend when needed state-of-the-art multi-class classification methods, such as SVM, kNN, prototype-based, logistic regression and ensemble methods, to the partial label learning framework. Moreover, we study the effect of incorporating the structure of the label set into the methods. We focus particularly on the hierarchical structure of the labels, as commonly observed in developmental processes. We show, on simulated and real datasets, that these extensions enable to learn from partially labeled data, and perform predictions with high accuracy, particularly with a nonlinear prototype-based method. We demonstrate that the performances of our methods trained with partially annotated data reach the same performance as fully supervised data. Finally, we study the level of uncertainty present in the partially annotated data, and derive some prescriptive results on the effect of this uncertainty on the accuracy of the partial label learning methods. Overall our findings show how hierarchical and non-hierarchical partial label learning strategies can help solve the problem of automated classification of single-cell transcriptomic profiles, interestingly these methods rely on a much less stringent type of annotated datasets compared to fully supervised learning methods.

Partial Label Learning with a Partner

In partial label learning (PLL), each instance is associated with a set of candidate labels among which only one is ground-truth. The majority of the existing works focuses on constructing robust classifiers to estimate the labeling confidence of candidate labels in order to identify the correct one. However, these methods usually struggle to rectify mislabeled samples. To help existing PLL methods identify and rectify mislabeled samples, in this paper, we introduce a novel partner classifier and propose a novel ``mutual supervision'' paradigm. Specifically, we instantiate the partner classifier predicated on the implicit fact that non-candidate labels of a sample should not be assigned to it, which is inherently accurate and has not been fully investigated in PLL. Furthermore, a novel collaborative term is formulated to link the base classifier and the partner one. During each stage of mutual supervision, both classifiers will blur each other's predictions through a blurring mechanism to prevent overconfidence in a specific label. Extensive experiments demonstrate that the performance and disambiguation ability of several well-established stand-alone and deep-learning based PLL approaches can be significantly improved by coupling with this learning paradigm.

Navigating Real-World Partial Label Learning: Unveiling Fine-Grained Images with Attributes

Partial label learning (PLL), a significant research area, addresses the challenge of annotating each sample with a candidate label set containing the true label when obtaining accurate labels is infeasible. However, existing PLL methods often rely on generic datasets like CIFAR, where annotators can readily differentiate candidate labels and are unlikely to confuse, making it less realistic for real-world partial label applications. In response, our research focuses on a rarely studied problem, PLL on fine-grained images with attributes. And we propose a novel framework called Shared to Learn, Distinct to Disambiguate (SoDisam). Within the candidate label set, the categories may exhibit numerous shared attribute features, posing a challenge in accurately distinguishing them. Rather than perceiving it as an impediment, we capitalize on these shared attributes as definitive sources of supervision. This insight guides us to learn attribute space visual representation to focus on the information from these shared attributes. Moreover, we introduce an attribute attention mechanism tailored to harness the remaining distinct attributes. This mechanism directs the originally holistic feature towards specific regions, capturing corresponding discriminative features. In addition, a dynamic disambiguation module is introduced, continuously adjusting the two aforementioned mechanisms and achieve the final disambiguation process. Extensive experiments demonstrate the effectiveness of our approach on fine-grained partial label datasets. The proposed SoDisam framework not only addresses the challenges associated with fine-grained partial label learning but also provides a more realistic representation of real-world partial label scenarios.

Federated Partial Label Learning with Local-Adaptive Augmentation and Regularization

Partial label learning (PLL) expands the applicability of supervised machine learning models by enabling effective learning from weakly annotated overcomplete labels. Existing PLL methods however focus on the standard centralized learning scenarios. In this paper, we expand PLL into the distributed computation setting by formalizing a new learning scenario named as federated partial label learning (FedPLL), where the training data with partial labels are distributed across multiple local clients with privacy constraints. To address this challenging problem, we propose a novel Federated PLL method with Local-Adaptive Augmentation and Regularization (FedPLL-LAAR). In addition to alleviating the partial label noise with moving-average label disambiguation, the proposed method performs MixUp-based local-adaptive data augmentation to mitigate the challenge posed by insufficient and imprecisely annotated local data, and dynamically incorporates the guidance of global model to minimize client drift through adaptive gradient alignment regularization between the global and local models. Extensive experiments conducted on multiple datasets under the FedPLL setting demonstrate the effectiveness of the proposed FedPLL-LAAR method for federated partial label learning.

Disentangled Partial Label Learning

Partial label learning (PLL) induces a multi-class classifier from training examples each associated with a set of candidate labels, among which only one is valid. The formation of real-world data typically arises from heterogeneous entanglement of series latent explanatory factors, which are considered intrinsic properties for discriminating between different patterns. Though learning disentangled representation is expected to facilitate label disambiguation for partial-label (PL) examples, few existing works were dedicated to addressing this issue. In this paper, we make the first attempt towards disentangled PLL and propose a novel approach named TERIAL, which makes predictions according to derived disentangled representation of instances and label embeddings. The TERIAL approach formulates the PL examples as an undirected bipartite graph where instances are only connected with their candidate labels, and employs a tailored neighborhood routing mechanism to yield disentangled representation of nodes in the graph. Specifically, the proposed routing mechanism progressively infers the explanatory factors that contribute to the edge between adjacent nodes and augments the representation of the central node with factor-aware embedding information propagated from specific neighbors simultaneously via iteratively analyzing the promising subspace clusters formed by the node and its neighbors. The estimated labeling confidence matrix is also introduced to accommodate unreliable links owing to the inherent ambiguity of PLL. Moreover, we theoretically prove that the neighborhood routing mechanism will converge to the point estimate that maximizes the marginal likelihood of observed PL training examples. Comprehensive experiments over various datasets demonstrate that our approach outperforms the state-of-the-art counterparts.

Controller-Guided Partial Label Consistency Regularization with Unlabeled Data

Partial label learning (PLL) learns from training examples each associated with multiple candidate labels, among which only one is valid. In recent years, benefiting from the strong capability of dealing with ambiguous supervision and the impetus of modern data augmentation methods, consistency regularization-based PLL methods have achieved a series of successes and become mainstream. However, as the partial annotation becomes insufficient, their performances drop significantly. In this paper, we leverage easily accessible unlabeled examples to facilitate the partial label consistency regularization. In addition to a partial supervised loss, our method performs a controller-guided consistency regularization at both the label-level and representation-level with the help of unlabeled data. To minimize the disadvantages of insufficient capabilities of the initial supervised model, we use the controller to estimate the confidence of each current prediction to guide the subsequent consistency regularization. Furthermore, we dynamically adjust the confidence thresholds so that the number of samples of each class participating in consistency regularization remains roughly equal to alleviate the problem of class-imbalance. Experiments show that our method achieves satisfactory performances in more practical situations, and its modules can be applied to existing PLL methods to enhance their capabilities.

Adaptive Integration of Partial Label Learning and Negative Learning for Enhanced Noisy Label Learning

There has been significant attention devoted to the effectiveness of various domains, such as semi-supervised learning, contrastive learning, and meta-learning, in enhancing the performance of methods for noisy label learning (NLL) tasks. However, most existing methods still depend on prior assumptions regarding clean samples amidst different sources of noise (e.g., a pre-defined drop rate or a small subset of clean samples). In this paper, we propose a simple yet powerful idea called NPN, which revolutionizes Noisy label learning by integrating Partial label learning (PLL) and Negative learning (NL). Toward this goal, we initially decompose the given label space adaptively into the candidate and complementary labels, thereby establishing the conditions for PLL and NL. We propose two adaptive data-driven paradigms of label disambiguation for PLL: hard disambiguation and soft disambiguation. Furthermore, we generate reliable complementary labels using all non-candidate labels for NL to enhance model robustness through indirect supervision. To maintain label reliability during the later stage of model training, we introduce a consistency regularization term that encourages agreement between the outputs of multiple augmentations. Experiments conducted on both synthetically corrupted and real-world noisy datasets demonstrate the superiority of NPN compared to other state-of-the-art (SOTA) methods. The source code has been made available at https://github.com/NUST-Machine-Intelligence-Laboratory/NPN.