False Labels Research Articles

Attention-disentangled re-ID network for unsupervised domain adaptive person re-identification

Unsupervised domain adaptation (UDA) for person re-identification (re-ID) aims to bridge the domain gap by transferring knowledge from the labeled source domain to the unlabeled target domain. Recently, pseudo-label-based approaches have become the dominant solution for addressing this issue. However,most pseudo-label-based methods neglect class boundary samples to reduce the influence of false pseudo labels, sacrificing intra-class semantic diversity. Although hard sample memory bank-based approaches can discover relationships between samples and describe intra-class diversity, they are prone to increasing the risks of generating incorrect pseudo labels. Balancing intra-class variety and pseudo-label accuracy in cross-domain person re-ID poses a challenge. In this study, we propose an attention-disentangled re-ID network (ADDNet) to enhance the discriminative ability of re-ID related feature representations, addressing the contradiction between intra-class diversity and pseudo-label accuracy. Unlike most disentanglement learning-based re-ID approaches that focus on separating explanatory factors, we design a spatial attention-disentangled mechanism to separate re-ID related and unrelated weights, enhancing the discriminative ability of cross-domain feature representation. Additionally, a multiple hard sample memory learning strategy is designed to express intra-class diversity of target samples using re-ID related features extracted from ADDNet. Extensive experiments show that ADDNet achieves 46.7% mAP on the Market-to-MSMT cross-domain re-ID task, outperforming state-of-the-art methods by 6.5 points.

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.

Multi-instance learning in the presence of positive and unlabeled bags

Multi-instance learning is a classic algorithm within the weakly supervised learning framework. Most previous approaches to multi-instance learning have typically assumed that every bag comes with its own label. However, in real-world applications, obtaining fully-labeled bags can be a challenging and resource-intensive task, primarily due to the substantial time and labor required for labeling each instance. Fortunately, collecting bags with only positive and unlabeled instances is a more feasible option. In this paper, we aim to learn an efficient binary classifier using positive and unlabeled bags. To address the absence of negative bags, we adopt a novel strategy that combines label noise learning and multi-instance kernels. Our approach involves a comprehensive analysis of empirical risk minimization in the presence of label noise, allowing us to decompose the loss function for negative bags with false negative labels into two distinct parts. Importantly, only the second part of this loss function is affected by the presence of noisy labels. To mitigate the influence of noisy labels on the negative bags, we propose the Multi-Instance Kernel-based Centroid Estimation (MIKCE) technique. MIKCE is a versatile method that can be employed with various types of data, including linear and nonlinear cases. Additionally, we provide an upper bound on the generalization error, ensuring the convergence of the MIKCE algorithm. Finally, we conduct numerical experiments on 31 public datasets to empirically validate the effectiveness of the MIKCE approach.

A Semi-Supervised Active Learning Method for Structured Data Enhancement with Small Samples

In order to solve the problems of the small capacity of structured data and uneven distribution among classes in machine learning tasks, a supervised generation method for structured data called WAGAN and a cyclic sampling method named SACS (Semi-supervised and Active-learning Cyclic Sampling), based on semi-supervised active learning, are proposed. The loss function and neural network structure are optimized, and the quantity and quality of the small sample set are enhanced. To enhance the reliability of generating pseudo-labels, a Semi-supervised Active learning Framework (SAF) is designed. This framework redistributes class labels to samples, which not only enhances the reliability of generated samples but also reduces the influence of noise and uncertainty on the generation of false labels. To mine the diversity information of generated samples, an uncertain sampling strategy based on spatial overlap is designed. This strategy incorporates the idea of spatial overlap and uses global and local sampling methods to calculate the information content of generated samples. Experimental results show that the proposed method performs better than other data enhancement methods on three different datasets. Compared to the original data, the average F1macro value of the classification model is improved by 11.5%, 16.1%, and 19.6% relative to compared methods.

Regularized joint self-training: A cross-domain generalization method for image classification

Unsupervised domain adaptive (UDA) is a widely used approach in machine learning strategies that helps move information from a source domain with rich supervised data to a target domain with limited supervised data. In recent years, self-training derived from semi-supervised learning provides a powerful tool for unsupervised domain adaptive by training unlabeled target domain data using pseudo-tags. However, in the case of domain shift, the pseudo-labels generated by standard self-training are no longer reliable due to the influence of the class-distribution shift in a specific domain, resulting in the noise of the pseudo-label error blocking the upper bound of the accuracy. To solve these problems, a domain adaptive method based on regularization joint self-training (RJS) is proposed in this paper. RJS consists of two key steps: First, the generation and optimization of pseudo-tags are trained jointly on the source domain to achieve automatic cross-domain generalization of pseudo-tags. Secondly, the regularization method is introduced as a confidence rule for false labels in initial model training, which implicitly prevents the model from remembering false labels. To verify the effectiveness of the proposed method, experiments were performed on VisDA-2017, Office-31, Office-Home, and DomainNet datasets, and the accuracy rates reached 87.1%, 89.3%, 76.9%, and 57.3%, respectively. In addition, the average classification performance of RJS on VisDA-2017 improved by about 12.6% compared to FixMatch, the standard self-training method. Experimental results demonstrate that RJS achieves state-of-the-art performance on multiple standard UDA benchmarks and confirm its effectiveness in the presence of domain-specific class distribution shifts.

AUTOMATIC IDENTIFICATION OF ISOLATED CALCIFICATIONS AND THEIR ACCUMULATIONS ON MAMMOGRAMS

Although microcalcinates usually are hyperattenuated, which makes them hyperintensive on mammograms, breast cancer is characterized by their small size, which, combined with the small size of their clusters, makes it difficult to identify them, especially against a dense background, which is often noted in fibrous changes in the breast parenchyma. The purpose of the study is to create and evaluate the effectiveness of the block for automatic identification of calcifications and their accumulations on mammograms. Material and methods. Mammograms of patients with suspicious (136 mammograms of 67 patients), as well as benign (299 mammograms of 151 patients) calcifications of various types were analyzed using a proprietary software package. Research results. After analyzing benign calcifications, the system marked all cases (100%) of calcified sediment, rod-shaped, vascular calcifications; 33 out of 36 (92.7%) cases of dystrophic, 66 out of 70 cases (94.3%) of rounded and 12 out of 15 (80%) cases of point calcifications in all patients; as well as 2 out of 3 cases of cutaneous calcifications in 1 out of 2 patients, 103 out of 106 (97.2%) cases of flaky calcifications in 51 out of 52 (98.1%) patients and 19 out of 22 cases (86.4%) of eggshell type calcifications in 10 out of 11 patients (90.9%). Among suspicious calcifications, the system marked 33 out of 39 cases (84.6%) of large heterogeneous calcifications, all 6 cases of small linear branching calcifications and 37 out of 39 (94.9%) cases of small polymorphic calcifications in all patients, as well as 30 out of 36 (83.3%) cases of amorphous calcifications in 15 out of 16 (93.7%) patients and 12 out of 16 (75.0%) cases of small linear calcifications in 6 out of 8 (75.0%) patients. All cases of unmarked suspicious calcifications corresponded to high-intensity soft-tissue shadows associated with indistinctly defined calcifications, which were labeled by the MammCheck II system previously developed by the authors. The frequency of false positive labels was 0.31 per mammogram. Conclusions. Benign calcifications were marked on 282 out of 299 images (94.3%) in 148 out of 151 (98.0%) patients, suspicious calcifications – on 118 out of 136 images (86.8%) in 64 out of 67 patients (95.5%).

Local–global pseudo-label correction for source-free domain adaptive medical image segmentation

Domain shift is a commonly encountered issue in medical imaging solutions, primarily caused by variations in imaging devices and data sources. To mitigate this problem, unsupervised domain adaptation techniques have been employed. However, concerns regarding patient privacy and potential degradation of image quality have led to an increased focus on source-free domain adaptation. In this study, we address the issue of false labels in self-training based source-free domain adaptive medical image segmentation methods. To correct erroneous pseudo-labels, we propose a novel approach called the local–global pseudo-label correction (LGDA) method for source-free domain adaptive medical image segmentation. Our method consists of two components: An offline local context-based pseudo-label correction method that utilizes local context similarity in image space. And an online global pseudo-label correction method based on class prototypes, which corrects erroneously predicted pseudo-labels by considering the relative distance between pixel-wise feature vectors and prototype vectors. We evaluate the performance of our method on three benchmark fundus image datasets for optic disc and cup segmentation. Our method achieves superior performance compared to the state-of-the-art approaches, even without using any source data.

WAS IT A MATch I SAW? Approximate palindromes lead to overstated false match rates in benchmarks using reversed sequences.

Software for labeling biological sequences typically produces a theory-based statistic for each match (the E-value) that indicates the likelihood of seeing that match's score by chance. E-values accurately predict false match rate for comparisons of random (shuffled) sequences, and thus provide a reasoned mechanism for setting score thresholds that enable high sensitivity with low expected false match rate. This threshold-setting strategy is challenged by real biological sequences, which contain regions of local repetition and low sequence complexity that cause excess matches between non-homologous sequences. Knowing this, tool developers often develop benchmarks that use realistic-seeming decoy sequences to explore empirical tradeoffs between sensitivity and false match rate. A recent trend has been to employ reversed biological sequences as realistic decoys, because these preserve the distribution of letters and the existence of local repeats, while disrupting the original sequence's functional properties. However, we and others have observed that sequences appear to produce high scoring alignments to their reversals with surprising frequency, leading to overstatement of false match risk that may negatively affect downstream analysis. We demonstrate that an alignment between a sequence S and its (possibly mutated) reversal tends to produce higher scores than alignment between truly unrelated sequences, even when S is a shuffled string with no notable repetitive or low-complexity regions. This phenomenon is due to the unintuitive fact that (even randomly shuffled) sequences contain palindromes that are on average longer than the longest common substrings (LCS) shared between permuted variants of the same sequence. Though the expected palindrome length is only slightly larger than the expected LCS, the distribution of alignment scores involving reversed sequences is strongly right-shifted, leading to greatly increased frequency of high-scoring alignments to reversed sequences. Overestimates of false match risk can motivate unnecessarily high score thresholds, leading to potentially reduced true match sensitivity. Also, when tool sensitivity is only reported up to the score of the first matched decoy sequence, a large decoy set consisting of reversed sequences can obscure sensitivity differences between tools. As a result of these observations, we advise that reversed biological sequences be used as decoys only when care is taken to remove positive matches in the original (un-reversed) sequences, or when overstatement of false labeling is not a concern. Though the primary focus of the analysis is on sequence annotation, we also demonstrate that the prevalence of internal palindromes may lead to an overstatement of the rate of false labels in protein identification with mass spectrometry.

MA-ST3D: Motion Associated Self-Training for Unsupervised Domain Adaptation on 3D Object Detection.

Recently, unsupervised domain adaptation (UDA) for 3D object detectors has increasingly garnered attention as a method to eliminate the prohibitive costs associated with generating extensive 3D annotations, which are crucial for effective model training. Self-training (ST) has emerged as a simple and effective technique for UDA. The major issue involved in ST-UDA for 3D object detection is refining the imprecise predictions caused by domain shift and generating accurate pseudo labels as supervisory signals. This study presents a novel ST-UDA framework to generate high-quality pseudo labels by associating predictions of 3D point cloud sequences during ego-motion according to spatial and temporal consistency, named motion-associated self-training for 3D object detection (MA-ST3D). MA-ST3D maintains a global-local pathway (GLP) architecture to generate high-quality pseudo-labels by leveraging both intra-frame and inter-frame consistencies along the spatial dimension of the LiDAR's ego-motion. It also equips two memory modules for both global and local pathways, called global memory and local memory, to suppress the temporal fluctuation of pseudo-labels during self-training iterations. In addition, a motion-aware loss is introduced to impose discriminated regulations on pseudo labels with different motion statuses, which mitigates the harmful spread of false positive pseudo labels. Finally, our method is evaluated on three representative domain adaptation tasks on authoritative 3D benchmark datasets (i.e. Waymo, Kitti, and nuScenes). MA-ST3D achieved SOTA performance on all evaluated UDA settings and even surpassed the weakly supervised DA methods on the Kitti and NuScenes object detection benchmark.

Multi-Label Chest X-Ray Image Classification with Single Positive Labels.

Deep learning approaches for multi-label Chest X-ray (CXR) images classification usually require large-scale datasets. However, acquiring such datasets with full annotations is costly, time-consuming, and prone to noisy labels. Therefore, we introduce a weakly supervised learning problem called Single Positive Multi-label Learning (SPML) into CXR images classification (abbreviated as SPML-CXR), in which only one positive label is annotated per image. A simple solution to SPML-CXR problem is to assume that all the unannotated pathological labels are negative, however, it might introduce false negative labels and decrease the model performance. To this end, we present a Multi-level Pseudo-label Consistency (MPC) framework for SPML-CXR. First, inspired by the pseudo-labeling and consistency regularization in semi-supervised learning, we construct a weak-to-strong consistency framework, where the model prediction on weakly-augmented image is treated as the pseudo label for supervising the model prediction on a strongly-augmented version of the same image, and define an Image-level Perturbation-based Consistency (IPC) regularization to recover the potential mislabeled positive labels. Besides, we incorporate Random Elastic Deformation (RED) as an additional strong augmentation to enhance the perturbation. Second, aiming to expand the perturbation space, we design a perturbation stream to the consistency framework at the feature-level and introduce a Feature-level Perturbation-based Consistency (FPC) regularization as a supplement. Third, we design a Transformer-based encoder module to explore the sample relationship within each mini-batch by a Batch-level Transformer-based Correlation (BTC) regularization. Extensive experiments on the CheXpert and MIMIC-CXR datasets have shown the effectiveness of our MPC framework for solving the SPML-CXR problem.

A study on fire data augmentation from video/image using the similar-label and F-guessed method

When data collection is limited, such as in the case of fire detection, improving the detection rate with only number of small labeled data is difficult. Therefore, researchers have conducted many related studies, among which semisupervised learning methods have achieved good results in improving detection rates. Most recent semi-supervised learning models use the pseudo-label method. But there is a problem, which is that it is difficult to label accurately in samples that deviate from the true label distribution due to false labels. In other words, due to the pseudo-label used for data augmentation, erroneous biases can be accumulated and adversely affect the final weights. To improve this, we proposed a method of generating Similar-labeled data (prediction result labeling value and correct answer value are similar), which was used through the F-guessed method and the Region of Interest (ROI) expression method in the video during initial learning. This has the effect of preventing the bias from being distorted in the initial stages. As a result, data generation increased by about 6.5 times, from 5,565 to 41,712, mAP@0.5 increased by about 26.1%, from 65.9% to 92.0%, and loss improved from 3.347 to 1.69, compared to the initial labeled data.

Food adulteration: Causes, risks, and detection techniques-review.

Food adulteration is the intentional addition of foreign or inferior substances to original food products for a variety of reasons. It takes place in a variety of forms, like mixing, substitution, hiding poor quality in packaging material, putting decomposed food for sale, misbranding or giving false labels, and adding toxicants. Several analytical methods (such as chromatography, spectroscopy, electronic sensors) are used to detect the quality of foodstuffs. This review provides concise but detailed information to understand the scope and scale of food adulteration as a way to further detect, combat, and prevent future adulterations. The objective of this review was to provide a comprehensive overview of the causes, risks, and detection techniques associated with food adulteration. It also aimed to highlight the potential health risks posed by consuming adulterated food products and the importance of detecting and preventing such practices. During the review, books, regulatory guidelines, articles, and reports on food adulteration were analyzed critically. Furthermore, the review assessed key findings to present a well-rounded analysis of the challenges and opportunities associated with combating food adulteration. This review included different causes and health impacts of food adulteration. The analytical techniques for food adulteration detection have also been documented in brief. In addition, the review emphasized the urgency of addressing food adulteration through a combination of regulatory measures, technological advancements, and consumer awareness. In conclusion, food adulteration causes many diseases such as cancer, liver disease, cardiovascular disease, kidney disease, and nervous system-related diseases. So, ensuring food safety is the backbone of health and customer satisfaction. Strengthening regulations, taking legal enforcement action, enhancing testing, and quality control can prevent and mitigate the adulteration of food products. Moreover, proper law enforcement and regular inspection of food quality can bring about drastic changes.

A self-learning deep neural network for classification of breast histopathological images

The most effective and feasible method for treating cancer is early diagnosis of breast cancer. An appropriate software tool, known as computer-aided diagnosis, helps doctors identify or diagnose cancer in the early stages and more quickly, hence decreasing the mortality rate. Convolution neural networks (CNNs) have lately been utilized in medical image analysis, which can discover cancer cells or classify histopathology images by processing a large amounts of training data. In this study, we carried out transfer learning (TL) accompanied with a new self-learning algorithm for classification of breast cancer by processing the histopathological images. In this study, several pre-trained deep neural networks including, Inception V3 Net, VGG19, Alex net, ResNet-18, Google net, Shefflenet, Mobile net, Resnet 101, Inception ResnetV2 Net, and Squeeze net were examined. Using a self-learning technique for classification of sub-images instead of the whole image can increase the accuracy of cancer classification. In this technique, the main issue is that the true label of sub-images is unknown, and the classifier should be trained using noisy labels. Therefore, a hierarchal self-learning method is utilized to correct the false labels gradually. The main rule for correcting the false labels is designed based on some prior knowledge about the errors exist in the initial labels of sub-images. The proposed self-learning method reached the accuracy of 99.1%, after 4 level of label correction process accompanied with Inception-V3Net pretrained network. Overall, the use of self-learning DNNs for the classification of histopathological breast cancer images can improve the accuracy, efficiency, interpretability of the classification results and contribute to the development of personalized and precision medicine.

Diagnosis of atrial fibrillation based on unsupervised domain adaptation

In recent years, the proportion of the elderly in the society is continuously increasing. Cardiovascular disease is a big problem that puzzles the health of the elderly. Among them, atrial fibrillation is one of the most common arrhythmia diseases in recent years, which poses a great threat to human life safety. At the same time, deep learning has become a powerful tool for medical and healthcare applications due to its high accuracy and fast detection speed. The diagnosis of atrial fibrillation is based on electrocardiogram, ECG) timing signals. At present, the scale of the open ECG data set is limited, and a large amount of labeled ECG data is needed to build a high-precision diagnostic model. In this study, a two-channel network model and a feature queue technique are proposed. A high-quality classification diagnosis model of atrial fibrillation is obtained by unsupervised domain adaptive technique, which uses a small amount of labeled data and a large amount of unlabeled data for training. The research content of this paper includes the following aspects: 1) Build a dual-channel network model, which can analyze ECG signals from different feature dimensions. At the same time, the dual-channel output also improves the reliability of the model's pseudo-label in the adaptive training stage and the accuracy of the output in the testing stage. 2) Innovative feature queue technology including global centroid is proposed to participate in the process of domain discrepancy metric calculation, which can use a small amount of labeled data and a large amount of unlabeled data to achieve a more stable and rapid update of the network. 3) Improved and innovated the domain discrepancy metric function, and introduced an evaluation formula for the credibility of false labels to improve the learning efficiency of unlabeled data. Finally, the experimental results show that the proposed two-channel network model and the feature queue technique with global centroid can achieve a high generalization and high precision depth network model by training with a small amount of labeled data and a large amount of unlabeled data. 4) The proposed model achieved a precision of 95.12%, a recall of 95.36%, an accuracy of 98.05%, and an F1 score of 95.23% in the MIT-BIH Arrhythmia Database. In the MIT-BIH Atrial Fibrillation Database, the model achieved a precision of 98.9%, a recall of 99.03%, an accuracy of 99.13%, and an F1 score of 99.08%.

Partial-Label Regression

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.

Video-Audio Domain Generalization via Confounder Disentanglement

Existing video-audio understanding models are trained and evaluated in an intra-domain setting, facing performance degeneration in real-world applications where multiple domains and distribution shifts naturally exist. The key to video-audio domain generalization (VADG) lies in alleviating spurious correlations over multi-modal features. To achieve this goal, we resort to causal theory and attribute such correlation to confounders affecting both video-audio features and labels. We propose a DeVADG framework that conducts uni-modal and cross-modal deconfounding through back-door adjustment. DeVADG performs cross-modal disentanglement and obtains fine-grained confounders at both class-level and domain-level using half-sibling regression and unpaired domain transformation, which essentially identifies domain-variant factors and class-shared factors that cause spurious correlations between features and false labels. To promote VADG research, we collect a VADG-Action dataset for video-audio action recognition with over 5,000 video clips across four domains (e.g., cartoon and game) and ten action classes (e.g., cooking and riding). We conduct extensive experiments, i.e., multi-source DG, single-source DG, and qualitative analysis, validating the rationality of our causal analysis and the effectiveness of the DeVADG framework.

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

Camera-irrelevant Feature Invariance Learning for Unsupervised Person Re-identification

Unsupervised person re-identification (re-ID) aims to match pictures of target pedestrians from some other cameras without any labels. The existing methods usually adopt a cluster-based pipeline to generate false labels for training. However, because of the differences in camera configuration, taking pictures Angle, illumination intensity and pedestrian background, unlabeled pedestrian training samples captured by different cameras are hard to classify into a cluster. To solve this problem, the paper proposes a camera-irrelevant feature learning framework, which learns camera-irrelevant features and camera-irrelevant distances through network training and clustering. Specifically, during training, images of other camera styles are generated, which enriches cross-camera learning. During clustering, we propose a camera-irrelevant distance matrix, which reduces the inter-class variance caused by camera variance.

Can Machine Learning Be Better than Biased Readers?

Background: Training machine learning (ML) models in medical imaging requires large amounts of labeled data. To minimize labeling workload, it is common to divide training data among multiple readers for separate annotation without consensus and then combine the labeled data for training a ML model. This can lead to a biased training dataset and poor ML algorithm prediction performance. The purpose of this study is to determine if ML algorithms can overcome biases caused by multiple readers' labeling without consensus. Methods: This study used a publicly available chest X-ray dataset of pediatric pneumonia. As an analogy to a practical dataset without labeling consensus among multiple readers, random and systematic errors were artificially added to the dataset to generate biased data for a binary-class classification task. The Resnet18-based convolutional neural network (CNN) was used as a baseline model. A Resnet18 model with a regularization term added as a loss function was utilized to examine for improvement in the baseline model. Results: The effects of false positive labels, false negative labels, and random errors (5-25%) resulted in a loss of AUC (0-14%) when training a binary CNN classifier. The model with a regularized loss function improved the AUC (75-84%) over that of the baseline model (65-79%). Conclusion: This study indicated that it is possible for ML algorithms to overcome individual readers' biases when consensus is not available. It is recommended to use regularized loss functions when allocating annotation tasks to multiple readers as they are easy to implement and effective in mitigating biased labels.

Cross-Domain CSI-Based Activity Recognition via Iterative Soft Labeling and Domain Alignment

The development of wireless technology has triggered wireless sensing. Most WiFi sensing methods are data driven and learning based. They face two major drawbacks: 1) a large number of labeled samples are required to train the sensing model; and 2) the sensing model depends on the training environment and degrades dramatically in a different environment. To mitigate these problems, we propose a domain adaptation method to achieve environment robustness for channel-state-information-based activity recognition with sparsely labeled samples. The method, named cross-domain activity recognition (CDAR), consists of iterative soft labeling, domain alignment, and activity classification. To reduce the number of labeled samples, CDAR adopts dynamic time warping to measure the similarity between the samples, based on which the unlabeled samples are pseudo-labeled iteratively and progressively. To tolerate false labels, the pseudo-labels take the form of soft labels. To reduce the data distribution discrepancy, the domains are aligned by minimizing the intraclass distance and maximizing the interclass distance, using maximum mean discrepancy as the metric. The activities are finally classified by integrating convolutional neural network and bidirectional long short-term memory. Extensive experiments demonstrate the effectiveness of the method CDAR on activity recognition across people, locations, environmental dynamics, and rooms.