Partial label learning induces classifier from data with ambiguous supervision, where each instance is associated with a set of candidate labels but only one of which is valid. As a classic data preprocessing strategy, dimensionality reduction contributes to enhance the generalization capabilities of learning algorithms. Due to the ambiguity of supervision, existing works on partial label dimensionality reduction are confined to two separate stages: dimensionality reduction and partial label disambiguation. However, the decoupling of dimensionality reduction from partial label disambiguation can lead to severe performance degradation. In this paper, we present a novel approach called Wrapped Partial Label Dimensionality Reduction (WPLDR) to address this challenge. Specifically, WPLDR integrates the dimensionality reduction and partial label disambiguation within a unified framework, employing alternating optimization to concurrently perform dimensionality reduction and partial label disambiguation. WPLDR maximizes the interdependence between features in the embedded space and confidence-based label information, while simultaneously ensuring the manifold consistency between the embedded feature space and label space. Extensive experiments over a broad range of synthetic and real-world partial label data sets validate that the performance of well-established partial label learning algorithms can be significantly improved by the proposed WPLDR.

Partial label learning (PLL) refers to the classification task where each training instance is ambiguously annotated with a set of candidate labels. Despite substantial advancements in tackling this challenge, limited attention has been devoted to a more specific and realistic setting, denoted as instance-dependent partial label learning (IDPLL). Within this contex, the assignment of partial labels depends on the distinct features of individual instances, rather than being random. In this paper, we initiate an exploration into a self-distillation framework for this problem, driven by the proven effectiveness and stability of this framework. Nonetheless, a crucial shortfall is identified: the foundational assumption central to IDPLL, involving what we term as partial label knowledge stipulating that candidate labels should exhibit superior confidence compared to non-candidates, is not fully upheld within the distillation process. To address this challenge, we introduce DIRK, a novel distillation approach that leverages a rectification process to DIstill Reliable Knowledge, while concurrently preserves informative fine-grained label confidence. In addition, to harness the rectified confidence to its fullest potential, we propose a knowledge-based representation refinement module, seamlessly integrated into the DIRK framework. This module effectively transmits the essence of similarity knowledge from the label space to the feature space, thereby amplifying representation learning and subsequently engendering marked improvements in model performance. Experiments and analysis on multiple datasets validate the rationality and superiority of our proposed approach.

Although deep neural networks have achieved remarkable success, they often exhibit a significant deficiency in reliable uncertainty calibration. This paper focus on model calibratability, which assesses how amenable a model is to be well recalibrated post-hoc. We find that the widely used weight decay regularizer detrimentally affects model calibratability, subsequently leading to a decline in final calibration performance after post-hoc calibration. To identify the underlying causes leading to poor calibratability, we delve into the calibratability of intermediate features across the hidden layers. We observe a U-shaped trend in the calibratability of intermediate features from the bottom to the top layers, which indicates that over-compression of the top representation layers significantly hinders model calibratability. Based on the observations, this paper introduces a weak classifier hypothesis, i.e., given a weak classification head that has not been over-trained, the representation module can be better learned to produce more calibratable features. Consequently, we propose a progressively layer-peeled training (PLP) method to exploit this hypothesis, thereby enhancing model calibratability. Our comparative experiments show the effectiveness of our method, which improves model calibration and also yields competitive predictive performance.

Noisy labels are often encountered in datasets, but learning with them is challenging. Although natural discrepancies between clean and mislabeled samples in a noisy category exist, most techniques in this field still gather them indiscriminately, which leads to their performances being partially robust. In this paper, we reveal both empirically and theoretically that the learning robustness can be improved by assuming deep features with the same labels follow a student distribution, resulting in a more intuitive method called student loss. By embedding the student distribution and exploiting the sharpness of its curve, our method is naturally data-selective and can offer extra strength to resist mislabeled samples. This ability makes clean samples aggregate tightly in the center, while mislabeled samples scatter, even if they share the same label. Additionally, we employ the metric learning strategy and develop a large-margin student (LT) loss for better capability. It should be noted that our approach is the first work that adopts the prior probability assumption in feature representation to decrease the contributions of mislabeled samples. This strategy can enhance various losses to join the student loss family, even if they have been robust losses. Experiments demonstrate that our approach is more effective in inaccurate supervision. Enhanced LT losses significantly outperform various state-of-the-art methods in most cases. Even huge improvements of over 50% can be obtained under some conditions.

Multiple-instance learning (MIL) solves the problem where training instances are grouped in bags, and a binary (positive or negative) label is provided for each bag. Most of the existing MIL studies need fully labeled bags for training an effective classifier, while it could be quite hard to collect such data in many real-world scenarios, due to the high cost of data labeling process. Fortunately, unlike fully labeled data, triplet comparison data can be collected in a more accurate and human-friendly way. Therefore, in this article, we for the first time investigate MIL from only triplet comparison bags , where a triplet (X a , X b , X c ) contains the weak supervision information that bag X a is more similar to X b than to X c . To solve this problem, we propose to train a bag-level classifier by the empirical risk minimization framework and theoretically provide a generalization error bound. We also show that a convex formulation can be obtained only when specific convex binary losses such as the square loss and the double hinge loss are used. Extensive experiments validate that our proposed method significantly outperforms other baselines.

Numerous researches have proved that deep neural networks (DNNs) can fit almost everything even given data with noisy labels, and result in poor generalization performance. However, recent studies suggest that DNNs tend to gradually memorize the data, moving from correct data to mislabeled data. Inspired by this finding, we propose a novel method namedDynamic Loss Thresholding (DLT). During the training process, DLT records the loss value of each sample and calculates dynamic loss thresholds. Specifically, DLT compares the loss value of each sample with the current loss threshold. Samples with smaller losses can be considered as clean samples with higher probability and vice versa. Then, DLT discards the potentially corrupted labels and further leverages self-training semi-supervised learning techniques. Experiments on CIFAR-10/100, WebVision and Clothing1M demonstrate substantial improvements over recent state-of-the-art methods. In addition, we investigate two real-world problems. Firstly, we propose a novel approach to estimate the noise rates of datasets based on the loss difference between the early and late training stages of DNNs. Secondly, we explore the effect of hard samples (which are difficult to be distinguished) on the process of learning from noisy labels.

Partial label learning learns from instances with weak supervision, where each instance is associated with a set of candidate labels, among which only one is valid. Recently, dimensionality reduction has emerged as an effective preprocessing strategy to improve generalization performance. Existing approaches mainly tackle this problem through supervised or unsupervised dimensionality reduction. However, the former requires ground-truth labels, which are concealed in candidate label sets. Consequently, methods in this line may suffer from overfitting due to false positive labels in candidate label set. Conversely, the latter overlooks weakly supervised information in training instances, leading to performance degradation. In this paper, we propose an approach calledpartial label Dimensionality Reduction via Adaptive Weight (Draw)to leverage the strengths of Principal Component Analysis (PCA) and Linear Discriminant Analysis (LDA). Specifically, our approach tends to exploit unsupervised and data-driven nature of PCA to capture underlying structure of instances in initial stage. As the ground-truth label is gradually identified, our method increasingly relies on the discriminative ability of LDA to enhance the separation between different classes. Through extensive experiments on diverse partial label datasets, we validate that the proposed dimensionality reduction approach significantly improves classification performance of well-established partial label learning algorithms.

Partial-label learning is a popular weakly supervised learning setting that allows each training example to be annotated with a set of candidate labels. Previous studies on partial-label learning only focused on the classification setting where candidate labels are all discrete, which cannot handle continuous labels with real values. In this paper, we provide the first attempt to investigate partial-label regression, where each training example is annotated with a set of real-valued candidate labels. To solve this problem, we first propose a simple baseline method that takes the average loss incurred by candidate labels as the predictive loss. The drawback of this method lies in that the loss incurred by the true label may be overwhelmed by other false labels. To overcome this drawback, we propose an identification method that takes the least loss incurred by candidate labels as the predictive loss. We further improve it by proposing a progressive identification method to differentiate candidate labels using progressively updated weights for incurred losses. We prove that the latter two methods are model-consistent and provide convergence analysis showing the optimal parametric convergence rate. Our proposed methods are theoretically grounded and can be compatible with any models, optimizers, and losses. Experiments validate the effectiveness of our proposed methods.

By simply taking convex combinations between pairs of samples and their labels, mixup training has been shown to easily improve predictive accuracy. It has been recently found that models trained with mixup also perform well on uncertainty calibration. However, in this study, we found that mixup training usually makes models less calibratable than vanilla empirical risk minimization, which means that it would harm uncertainty estimation when post-hoc calibration is considered. By decomposing the mixup process into data transformation and random perturbation, we suggest that the confidence penalty nature of the data transformation is the reason of calibration degradation. To mitigate this problem, we first investigate the mixup inference strategy and found that despite it improves calibration on mixup, this ensemble-like strategy does not necessarily out-perform simple ensemble. Then, we propose a general strategy named mixup inference in training, which adopts a simple decoupling principle for recovering the outputs of raw samples at the end of forward network pass. By embedding the mixup inference, models can be learned from the original one-hot labels and hence avoid the negative impact of confidence penalty. Our experiments show this strategy properly solves mixup's calibration issue without sacrificing the predictive performance, while even improves accuracy than vanilla mixup.

Partial label learning (PLL), which refers to the classification task where each training instance is ambiguously annotated with a set of candidate labels, has been recently studied in deep learning paradigm. Despite advances in recent deep PLL literature, existing methods (e.g., methods based on self-training or contrastive learning) are confronted with either ineffectiveness or inefficiency. In this paper, we revisit a simple idea namely consistency regularization, which has been shown effective in traditional PLL literature, to guide the training of deep models. Towards this goal, a new regularized training framework, which performs supervised learning on non-candidate labels and employs consistency regularization on candidate labels, is proposed for PLL. We instantiate the regularization term by matching the outputs of multiple augmentations of an instance to a conformal label distribution, which can be adaptively inferred by the closed-form solution. Experiments on benchmark datasets demonstrate the superiority of the proposed method compared with other state-of-the-art methods.

In partial label learning, a multi-class classifier is learned from the ambiguous supervision where each training example is associated with a set of candidate labels among which only one is valid. An intuitive way to deal with this problem is label disambiguation, i.e., differentiating the labeling confidences of different candidate labels so as to try to recover ground-truth labeling information. Recently, feature-aware label disambiguation has been proposed which utilizes the graph structure of feature space to generate labeling confidences over candidate labels. Nevertheless, the existence of noises and outliers in training data makes the graph structure derived from original feature space less reliable. In this paper, a novel partial label learning approach based on adaptive graph guided disambiguation is proposed, which is shown to be more effective in revealing the intrinsic manifold structure among training examples. Other than the sequential disambiguation-then-induction learning strategy, the proposed approach jointly performs adaptive graph construction, candidate label disambiguation and predictive model induction with alternating optimization. Furthermore, we consider the particular human-in-the-loop framework in which a learner is allowed to actively query some ambiguously labeled examples for manual disambiguation. Extensive experiments clearly validate the effectiveness of adaptive graph guided disambiguation for learning from partial label examples.

Recent researches reveal that deep neural networks are sensitive to label noises hence leading to poor generalization performance in some tasks. Although different robust loss functions have been proposed to remedy this issue, they suffer from an underfitting problem, thus are not sufficient to learn accurate models. On the other hand, the commonly used Cross Entropy (CE) loss, which shows high performance in standard supervised learning (with clean supervision), is non-robust to label noise. In this paper, we propose a general framework to learn robust deep neural networks with complementary loss functions. In our framework, CE and robust loss play complementary roles in a joint learning objective as per their learning sufficiency and robustness properties respectively. Specifically, we find that by exploiting the memorization effect of neural networks, we can easily filter out a proportion of hard samples and generate reliable pseudo labels for easy samples, and thus reduce the label noise to a quite low level. Then, we simply learn with CE on pseudo supervision and robust loss on original noisy supervision. In this procedure, CE can guarantee the sufficiency of optimization while the robust loss can be regarded as the supplement. Experimental results on benchmark classification datasets indicate that the proposed method helps achieve robust and sufficient deep neural network training simultaneously.

In complementary-label learning (CLL), a multi-class classifier is learned from training instances each associated with complementary labels, which specify the classes that the instance does not belong to. Previous studies focus on unbiased risk estimator or surrogate loss while neglect the importance of regularization in training phase. In this paper, we give the first attempt to leverage regularization techniques for CLL. By decoupling a label vector into complementary labels and partial unknown labels, we simultaneously inhibit the outputs of complementary labels with a complementary loss and penalize the sensitivity of the classifier on the partial outputs of these unknown classes by consistency regularization. Then we unify the complementary loss and consistency loss together by a specially designed dynamic weighting factor. We conduct a series of experiments showing that the proposed method achieves highly competitive performance in CLL.

Capturing accurate uncertainty quantification of the prediction from deep neural networks is important in many real-world decision-making applications. A reliable predictor is expected to be accurate when it is confident about its predictions and indicate high uncertainty when it is likely to be inaccurate. However, modern neural networks have been found to be poorly calibrated, primarily in the direction of overconfidence. In recent years, there is a surge of research on model calibration by leveraging implicit or explicit regularization techniques during training, which obtain well calibration by avoiding overconfident outputs. In our study, we empirically found that despite the predictions obtained from these regularized models are better calibrated, they suffer from not being as calibratable, namely, it is harder to further calibrate their predictions with post-hoc calibration methods like temperature scaling and histogram binning. We conduct a series of empirical studies showing that overconfidence may not hurt final calibration performance if post-hoc calibration is allowed, rather, the penalty of confident outputs will compress the room of potential improvements in post-hoc calibration phase. Our experimental findings point out a new direction to improve calibration of DNNs by considering main training and post-hoc calibration as a unified framework.

Multi-view multi-label learning serves an important framework to learn from objects with diverse representations and rich semantics. Existing multi-view multi-label learning techniques focus on exploiting shared subspace for fusing multi-view representations, where helpful view-specific information for discriminative modeling is usually ignored. In this paper, a novel multi-view multi-label learning approach named SIMM is proposed which leverages shared subspace exploitation and view-specific information extraction. For shared subspace exploitation, SIMM jointly minimizes confusion adversarial loss and multi-label loss to utilize shared information from all views. For view-specific information extraction, SIMM enforces an orthogonal constraint w.r.t. the shared subspace to utilize view-specific discriminative information. Extensive experiments on real-world data sets clearly show the favorable performance of SIMM against other state-of-the-art multi-view multi-label learning approaches.

Partial label learning aims to induce a multi-class classifier from training examples where each of them is associated with a set of candidate labels, among which only one is the ground-truth label. The common strategy to train predictive model is disambiguation, i.e. differentiating the modeling outputs of individual candidate labels so as to recover ground-truth labeling information. Recently, feature-aware disambiguation was proposed to generate different labeling confidences over candidate label set by utilizing the graph structure of feature space. However, the existence of noise and outliers in training data makes the similarity derived from original features less reliable. To this end, we proposed a novel approach for partial label learning based on adaptive graph guided disambiguation (PL-AGGD). Compared with fixed graph, adaptive graph could be more robust and accurate to reveal the intrinsic manifold structure within the data. Moreover, instead of the two-stage strategy in previous algorithms, our approach performs label disambiguation and predictive model training simultaneously. Specifically, we present a unified framework which jointly optimizes the ground-truth labeling confidences, similarity graph and model parameters to achieve strong generalization performance. Extensive experiments show that PL-AGGD performs favorably against state-of-the-art partial label learning approaches.

Traffic flow prediction is an important part of intelligent transportation systems (ITS). However, the performance of current traffic flow prediction methods does not meet the expectation. Weather factors such as precipitation in residential areas and tourist destinations affect traffic flow on the surrounding roads. In this paper, we attempt to take precipitation impact into consideration when predicting traffic flow. To realize this idea, we propose a deep traffic flow prediction architecture by introducing a deep bi-directional long short-term memory model, precipitation information, residual connection, regression layer and dropout training method. The proposed model has good ability to capture the deep features of traffic flow. Besides, it can take full advantage of time-aware traffic flow data and additional precipitation data. We evaluate the prediction architecture on the dataset from Caltrans Performance Measurement System (PeMS) with precipitation data from California Data Exchange Center (CDEC) and the dataset from KDD Cup 2017. The experiment results demonstrate that the proposed model for traffic flow prediction obtains high accuracy and generalizes well compared with other models.

With the overload of information on the Web, Recommender Systems (RSs) are becoming increasingly popular and have been employed to provide suggestions to meet different requirements. RSs are utilized in a variety of areas including movies, music, social tags, user group and products as Web services evoked on the Internet either as mobile Apps or PC-based applications. However, it is challenging to achieve personalized recommendations instead of offering up too many lowest common denominator recommendations. Understanding how products relate to each other is important because it has great impact on the performance. Furthermore, the personalized sequential behavior, which is closely related to a particular product, is essential for recommender systems. Most models simply integrate features from users and items without considering potential product bundle relationships between products exposed by users' personalized sequential behaviors. In this paper, a novel method based on Factorizing Personalized Markov Chain (FPMC) is proposed to comprehensively explore the latent bundle relations from users perspective, along with the hidden correlative semantics between products obtained from logic regression method, which provides a unified view to describe the user preferences, product/item features, and the user sequential patterns in timely manner. The involved semantic features are extracted using deep learning models. We evaluate our method on real-world Amazon datasets and our framework significantly outperforms other baseline models, especially on sparse datasets. The experimental results show that our approach qualitatively captures personalized behaviors with superior recommendation performance.