Instance-Dependent Noisy Label Learning via Graphical Modelling

Point cloud segmentation is a fundamental task in 3D. Despite recent progress on point cloud segmentation with the power of deep networks, current learning methods based on the clean label assumptions may fail with noisy labels. Yet, class labels are often mislabeled at both instance-level and boundary-level in real-world datasets. In this work, we take the lead in solving the instance-level label noise by proposing a Point Noise-Adaptive Learning (PNAL) framework. Compared to noise-robust methods on image tasks, our framework is noise-rate blind, to cope with the spatially variant noise rate specific to point clouds. Specifically, we propose a point-wise confidence selection to obtain reliable labels from the historical predictions of each point. A cluster-wise label correction is proposed with a voting strategy to generate the best possible label by considering the neighbor correlations. To handle boundary-level label noise, we also propose a variant "PNAL-boundary" with a progressive boundary label cleaning strategy. Extensive experiments demonstrate its effectiveness on both synthetic and real-world noisy datasets. Even with 60% symmetric noise and high-level boundary noise, our framework significantly outperforms its baselines, and is comparable to the upper bound trained on completely clean data. Moreover, we cleaned the popular real-world dataset ScanNetV2 for rigorous experiment. Our code and data is available at https://github.com/pleaseconnectwifi/PNAL.

The prevalence of mobile sensors makes it possible for researchers to collect and analyze the pervasive sensed human activity data with machine learning tools. These analyses and applications heavily rely on the self-report survey data that reflect human physical and psychological behaviors. However, many factors can affect the reliability of such self-report surveys, such as participants' trusty and ambiguous descriptions in questionnaires. These inaccurate survey labels challenge the development of the models, especially when it comes to the case where deep learning approaches tend to memorize the noisy information from training data. In this paper, we set out to deal with the noisy self-report survey labels. We aim to address the following challenges: 1)the limited access to clean labeled data in real-world scenarios; 2)different noise ratio of the data from different participants; 3)various types of noise that appear in the data; 4)how to evaluate the reliability of participants. As such, we propose a generalizable model, Lan, for learning from multi-modal sensory data and noisy labels. The proposed model incorporates the temporal information from sensory data to guide the model training and thus to augment noise tolerance. Extensive experiments on a real-world dataset were conducted to show our model outperforms state-of-the-art benchmarks.

Mutual calibration training: Training deep neural networks with noisy labels using dual-models

Wearable technologies enable continuous monitoring of various health metrics, such as physical activity, heart rate, sleep, and stress levels. A key challenge with wearable data is obtaining quality labels. Unlike modalities like video where the videos themselves can be effectively used to label objects or events, wearable data do not contain obvious cues about the physical manifestation of the users and usually require rich metadata. As a result, label noise can become an increasingly thorny issue when labeling wearable data. In this paper, we propose a novel solution to address noisy label learning, entitled Few-Shot Human-in-the-Loop Refinement (FHLR). Our method initially learns a seed model using weak labels. Next, it fine-tunes the seed model using a handful of expert corrections. Finally, it achieves better generalizability and robustness by merging the seed and fine-tuned models via weighted parameter averaging. We evaluate our approach on four challenging tasks and datasets, and compare it against eight competitive baselines designed to deal with noisy labels. We show that FHLR achieves significantly better performance when learning from noisy labels and achieves state-of-the-art by a large margin, with up to accuracy improvement under symmetric and asymmetric noise. Notably, we find that FHLR is particularly robust to increased label noise, unlike prior works that suffer from severe performance degradation. Our work not only achieves better generalization in high-stakes health sensing benchmarks but also sheds light on how noise affects commonly-used models.

The performance of network traffic detection heavily relies on high-quality annotated data, yet the widespread presence of noisy labels in real-world scenarios severely undermines model reliability and generalization. Existing methods predominantly rely on training dynamics signals and struggle to distinguish between noisy labels and valuable hard samples, leading to diminished model sensitivity to emerging threats. To address this fundamental challenge, this paper proposes a noise-tolerant label recognition and correction framework based on graph-structured neighbor consistency (NCRT). The framework leverages the inherent clustering characteristics of network traffic in feature space, constructs an adaptive K-nearest neighbor graph to model behavioral similarities among samples, and employs multi-dimensional consistency evaluation metrics to accurately differentiate between noisy labels and hard samples. Furthermore, a feature-space-based label propagation algorithm is introduced, which utilizes reliable anchor samples in the graph to guide noisy labels toward correction based on the consensus of their semantic neighbors. The core innovation lies in leveraging graph-structured neighbor consistency to fundamentally distinguish noisy labels from hard samples, and correcting them via label propagation over an adaptive K-nearest neighbor graph. Experiments on real-world datasets such as CICIDS2017 and NSL-KDD demonstrate that the proposed method significantly outperforms mainstream baseline approaches in terms of noise identification, label correction, and final classification performance. It exhibits exceptional robustness and stability, particularly under high-ratio and asymmetric noise settings, providing an effective solution for building reliable traffic detection models adapted to complex real-world network environments.

Network Intrusion Detection Systems (NIDSs) are an increasingly important tool for the prevention and mitigation of cyber attacks. Over the past years, a lot of research efforts have aimed at leveraging the increasingly powerful models of Machine Learning (ML) for this purpose. A number of labelled synthetic datasets have been generated and made publicly available by researchers, and they have become the benchmarks via which new ML-based NIDS classifiers are being evaluated. Recently published results show excellent classification performance with these datasets, increasingly approaching 100 percent performance across key evaluation metrics such as Accuracy, F1 score, AUC, etc. Unfortunately, we have not yet seen these excellent academic research results translated into practical NIDS systems with such near-perfect performance. This motivated our research presented in this paper, where we analyse the statistical properties of the benign traffic in three of the more recent and relevant NIDS datasets, (CIC_IDS, UNSW_NB15, TON_IOT), by converting them into a common flow format. As a comparison, we consider two datasets obtained from real-world production networks, one from a university network and one from a medium size Internet Service Provider (ISP). Our results show that the two real-world datasets are quite similar among themselves in regards to most of the considered statistical features. Equally, the three synthetic datasets are also relatively similar within their group. However, and most importantly, our results show a distinct difference of most of the considered statistical features between the three synthetic datasets and the two real-world datasets. Since ML relies on the basic assumption of training and test datasets being sampled from the same distribution, this raises the question of how well the performance results of ML-classifiers trained on the considered synthetic datasets can translate and generalise to real-world networks. We believe this is an interesting and relevant question which provides motivation for further research in this space.

As a pivotal task in 3D scene understanding, 3D point cloud semantic segmentation is often challenged by the presence of mislabeled data in real-world annotations. Such noisy labels can mislead the segmentation model into learning erroneous features during training, thereby degrading its performance. To address this issue, this paper proposes a robust training framework for 3D semantic segmentation based on noisy label correction. The proposed method does not rely on prior knowledge of point cloud density and is capable of learning robust segmentation models under various noise rates. Our approach consists of two core components: a Density-Adaptive Neighborhood Clustering algorithm that dynamically adjusts the neighborhood radius and minimum point count according to local point density variations, and a Dynamically-Weighted Label Refurbishing mechanism, which rectifies noisy labels at the cluster level based on historical prediction confidence. Extensive experiments on the S3DIS dataset demonstrate that our method outperforms existing baselines and state-of-the-art (SOTA) approaches under symmetric, asymmetric, and mixed noise settings. Notably, it maintains strong robustness even under high noise ratios. Furthermore, our method integrates label structures and stores them in a vectorized format, significantly reducing data transfer overhead between CPU and GPU and mitigating resource wastage. This design makes the approach particularly efficient in resource-constrained environments such as cloud servers, where CPU performance is often a bottleneck. As a result, our method achieves a notable improvement in training efficiency, with each training epoch requiring only approximately one-tenth of the time compared to current SOTA methods.

Traditional computational approaches to design chemical species are limited by the need to compute properties for a vast number of candidates, e.g., by discriminative modeling. Therefore, inverse design methods aim to start from the desired property and optimize a corresponding chemical structure. From a machine learning viewpoint, the inverse design problem can be addressed through so-called generative modeling. Mathematically, discriminative models are defined by learning the probability distribution function of properties given the molecular or material structure. In contrast, a generative model seeks to exploit the joint probability of a chemical species with target characteristics. The overarching idea of generative modeling is to implement a system that produces novel compounds that are expected to have a desired set of chemical features, effectively sidestepping issues found in the forward design process. In this contribution, we overview and critically analyze popular generative algorithms like generative adversarial networks, variational autoencoders, flow, and diffusion models. We highlight key differences between each of the models, provide insights into recent success stories, and discuss outstanding challenges for realizing generative modeling discovered solutions in chemical applications.

Real-world datasets often suffer from both noisy labels and imbalanced class distribution, presenting significant challenges for the effective deployment of deep neural networks (DNNs). Existing studies typically address these challenges separately and struggle to perform effectively when they occur simultaneously. In this paper, we introduce an unbiased Sample Selection method based on the Graph Attention Network (GAT), namely GSS. GSS can effectively divide the training set into clean and noisy subsets while avoiding sample selection bias by analyzing the intrinsic relationships between the training set and a small clean validation set. For the clean subset, we propose an Adaptive Label Refinement (ALR) strategy to improve the reliability of the labels within the clean subset. ALR dynamically integrates the network’s predictions with the given labels, mitigating the adverse impacts of misidentification. For the noisy subset, we introduce a Class-Balanced Pseudo Labeling (CBPL) method. CBPL addresses the cognitive bias in model predictions caused by class imbalance by integrating class distribution information into the pseudo-label generation process, resulting in more accurate pseudo-labels. Comprehensive evaluations on both synthetic and real-world datasets highlight the effectiveness and superiority of our approach, especially in scenarios characterized by noisy labels and imbalanced class distributions.

Hidden information derived from probabilistic generative models of data distributions can be used to construct features for discriminative classifiers. This observation has motivated the development of approaches that attempt to couple generative and discriminative models together for classification. However, existing approaches typically feed features derived from generative models to discriminative classifiers, and do not refine the generative models or the feature mapping functions based on classification results. In this paper, we propose a coupling mechanism developed under the PAC-Bayes framework that can fine-tune the generative models and the feature mapping functions iteratively to improve the classifier's performance. In our approach, a stochastic feature mapping, which is a function over the random variables of a generative model, is derived to generate feature vectors for a stochastic classifier. We construct a stochastic classifier over the feature mapping and derive the PAC-Bayes generalization bound for the classifier, for both supervised and semi-supervised learning. This allows us to jointly learn the feature mapping and the classifier by minimizing the bound with an EM-like iterative algorithm using labeled and unlabeled data. The resulting framework integrates the learning of the discriminative classifier and the generative model and allows iterative fine-tuning of the generative models, and the feedforward feature mappings based on task performance feedback. Our experiments show, in three distinct applications, this new framework produces a general classification tool with state-of-the-art performance.

Deep generative models have demonstrated an excellent ability to generate data by learning their distribution. Despite their unsupervised nature, these models can be implemented in semi-supervised learning scenarios by treating the class labels as additional latent variables. In this article, we propose a deep generative model for semi-supervised learning that offsets label noise, which is a ubiquitous feature in large-scale datasets owing to the high cost of annotation. We assume that noisy labels are generated from true labels and employ a noise transition matrix to describe the transition. We estimate this matrix by adjusting its entries to minimize its difference with the true transition matrix and use the estimated matrix to formulate the objective function for inference, which consists of an evidence lower bound and a classification risk. However, because directly minimizing the latter with noisy labels may result in an inaccurate classifier, we propose a statistically consistent estimator for computing the classification risk solely with noisy data. Empirical results on benchmark datasets demonstrate that the proposed model improves the classification performance over that of the baseline algorithms. We also present a case study on semiconductor manufacturing. Additionally, we empirically show that the proposed model, as a generative model, is capable of reconstructing data even with noisy labels.

Image classification serves as the cornerstone of computer vision, traditionally achieved through discriminative models based on deep neural networks. Recent advancements have introduced classification methods derived from generative models, which offer the advantage of zero-shot classification. However, these methods suffer from two main drawbacks: high computational overhead and inferior performance compared to discriminative models. Inspired by the coordinated cognitive processes of rapid-slow pathway interactions in the human brain during visual signal recognition, we propose the Diffusion-Based Discriminative Model Enhancement Framework (DBMEF). This framework seamlessly integrates discriminative and generative models in a training-free manner, leveraging discriminative models for initial predictions and endowing deep neural networks with rethinking capabilities via diffusion models. Consequently, DBMEF can effectively enhance the classification accuracy and generalization capability of discriminative models in a plug-and-play manner. We have conducted extensive experiments across 17 prevalent deep model architectures with different training methods, including both CNN-based models such as ResNet and Transformer-based models like ViT, to demonstrate the effectiveness of the proposed DBMEF.Specifically, the framework yields a 1.51% performance improvement for ResNet-50 on the ImageNet dataset and 3.02% on the ImageNet-A dataset. In conclusion, our research introduces a novel paradigm for image classification, demonstrating stable improvements across different datasets and neural networks.

Bottleneck problems are an important class of optimization problems that have recently gained increasing attention in the domain of machine learning and information theory. They are widely used in generative models, fair machine learning algorithms, design of privacy-assuring mechanisms, and appear as information-theoretic performance bounds in various multi-user communication problems. In this work, we propose a general family of optimization problems, termed as <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">complexity-leakage-utility bottleneck (CLUB)</i> model, which (i) provides a unified theoretical framework that generalizes most of the state-of-the-art literature for the information-theoretic privacy models, (ii) establishes a new interpretation of the popular generative and discriminative models, (iii) constructs new insights for the generative compression models, and (iv) can be used to obtain fair generative models. We first formulate the CLUB model as a complexity-constrained privacy-utility optimization problem. We then connect it with the closely related bottleneck problems, namely information bottleneck (IB), privacy funnel (PF), deterministic IB (DIB), conditional entropy bottleneck (CEB), and conditional PF (CPF). We show that the CLUB model generalizes all these problems as well as most other information-theoretic privacy models. Then, we construct the deep variational CLUB (DVCLUB) models by employing neural networks to parameterize variational approximations of the associated information quantities. Building upon these information quantities, we present unified objectives of the <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">supervised</i> and <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">unsupervised</i> DVCLUB models. Leveraging the DVCLUB model in an unsupervised setup, we then connect it with state-of-the-art generative models, such as variational auto-encoders (VAEs), generative adversarial networks (GANs), as well as the Wasserstein GAN (WGAN), Wasserstein auto-encoder (WAE), and adversarial auto-encoder (AAE) models through the optimal transport (OT) problem. We then show that the DVCLUB model can also be used in fair representation learning problems, where the goal is to mitigate the undesired bias during the training phase of a machine learning model. We conduct extensive quantitative experiments on colored-MNIST and CelebA datasets.

This paper presents a fusing framework of discriminative and generative models for Chinese dialect identification. The generative models are employed to produce language feature vectors and the discriminative models are used to make classification. Four Chinese dialects is tested with this system. The experimental results showed that the proposed system outperformed the GMM based system. Meanwhile the SVM based discriminative methods has more powerful discriminative ability than ANN based one.

Generative models offer advantageous characteristics for classification tasks, such as the availability of unsupervised data and calibrated confidence. In contrast, discriminative models have advantages in terms of their potential to outperform their generative counterparts and the simplicity of their model structures and learning algorithms. In this article, we propose a method to train a hybrid of discriminative and generative models in a single neural network (NN), which exhibits the characteristics of both models. The key idea is the Gaussian-coupled softmax layer, which is a fully connected layer with a softmax activation function coupled with Gaussian distributions. This layer can be embedded into an NN-based classifier and allows the classifier to estimate both the class posterior distribution and the input data distribution. We demonstrate that the proposed hybrid model can be applied to semi-supervised learning and confidence calibration.