Noisy Data Research Articles

Quantitative evaluation of intensity fidelity of super-resolution reconstruction for structured illumination microscopy

Super-resolution structured illumination microscopy (SR-SIM) relies heavily on post-processing reconstruction to obtain high-quality SR images from raw data. Although many SIM reconstruction algorithms have been developed to recover fine cellular structures with high fidelity even from the noisy data, whether the pixel intensities of reconstructed SR images are still proportional to the original fluorescence intensity has been less explored. The linearity between the intensity before and after reconstruction is defined as the intensity fidelity. Here, we proposed a method to evaluate the reconstructed SR image intensity fidelity at different spatial frequencies. With the proposed metric, we systematically investigated the impact of the key factors on the intensity fidelity in the standard Wiener-SIM reconstructions with simulated data, then evaluated the intensity fidelity of the SR images reconstructed by representative open-source packages. Our work provides a reference for SR-SIM image intensity fidelity improvement.

Improving EEG Forward Modeling Using High-Resolution Five-Layer BEM-FMM Head Models: Effect on Source Reconstruction Accuracy

Electroencephalographic (EEG) source localization is a fundamental tool for clinical diagnoses and brain-computer interfaces. We investigate the impact of model complexity on reconstruction accuracy by comparing the widely used three-layer boundary element method (BEM) as an inverse method against a five-layer BEM accelerated by the fast multipole method (BEM-FMM) and coupled with adaptive mesh refinement (AMR) as forward solver. Modern BEM-FMM with AMR can solve high-resolution multi-tissue models efficiently and accurately. We generated noiseless 256-channel EEG data from 15 subjects in the Connectome Young Adult dataset, using four anatomically relevant dipole positions, three conductivity sets, and two head segmentations; we mapped localization errors across the entire grey matter from 4,000 dipole positions. The average location error among our four selected dipoles is ∼5mm (±2mm) with an orientation error of ∼12∘ (±7∘). The average source localization error across the entire grey matter is ∼9mm (±4mm), with a tendency for smaller errors on the occipital lobe. Our findings indicate that while three-layer models are robust under noiseless conditions, substantial localization errors (10–20mm) are common. Therefore, models of five or more layers may be needed for accurate source reconstruction in critical applications involving noisy EEG data.

Enhanced robustness in machine learning: Application of an adaptive robust loss function

In the field of machine learning, the selection of a loss function plays a pivotal role in determining the training dynamics and generalization capability of models. Traditional loss functions such as mean square error (MSE) and cross-entropy are often not equipped to handle noisy and anomalous data effectively, which can result in models that perform poorly in practical scenarios. To address these shortcomings, this study introduces a novel adaptive robust loss function, enhanced by a tunable parameter , which allows for flexibility in adjusting the robustness of the function according to the nature of the data being processed. Our research demonstrates that this new loss function significantly improves the performance of linear regression and multilayer perceptron models, particularly in environments laden with noisy and anomalous data. By adapting the parameter , the function can cater to varying levels of data irregularities, thus enhancing the models accuracy and reliability across diverse and complex data environments. This adaptive mechanism not only offers a substantial theoretical contribution to the understanding of robust loss functions but also provides a practical tool for machine learning practitioners to develop models that are resilient to data imperfections. The implications of this research are profound, suggesting a shift towards more adaptive and robust approaches in machine learning model development.

Improvement for deep learning for suicide and depression identification with unsupervised label correction

Abstract. Our study try to address the challenge of accurately identifying depression and suicidal ideation on social media platforms by introducing an enhanced novel methodology for unsupervised feature selection and label correction. Utilizing advanced word embedding models like BERT and the Universal Sentence Encoder, we transform textual content into dense numerical vectors that capture the nuanced emotional context of online discussions. Our approach enhances these embeddings with a deep neural network (DNN) to extract distinctive features, reducing the dimensionality of the data through Principal Component Analysis (PCA). For label correction, we employ clustering techniques including OPTICS, K-medoids, and hierarchical clustering, which are robust against noisy data points. We then train classifiers using CNN, DNN, logistic regression, and random forest algorithms, evaluated with metrics such as accuracy, precision, recall, F1 score, and AUC. This methodology improves the accuracy of classifying depressive and suicidal sentiments to some extent, assisting to utilize the vast data available on social media to advance mental health diagnostics and interventions.

Data-driven computation of adjoint sensitivities without adjoint solvers: An application to thermoacoustics

Adjoint methods have been the pillar of gradient-based optimization for decades. They enable the accurate computation of a gradient (sensitivity) of a quantity of interest with respect to all system parameters in one calculation. When the gradient is embedded in an optimization routine, the quantity of interest can be optimized for the system to have the desired behavior. Adjoint methods, however, require the system's governing equations and their Jacobian. In this paper, we propose a computational strategy to infer the adjoint sensitivities from data when the governing equations might be unknown (or partly unknown), and noise might be present. The key component of this strategy is an echo state network, which learns the dynamics of nonlinear regimes with varying parameters, and it evolves dynamically via a hidden state. Although the framework does not make assumptions on the dynamical system, we focus on thermoacoustics, which are governed by nonlinear and time-delayed systems. First, we show that a parameter-aware echo state network (ESN) infers the parametrized dynamics. Second, we derive the adjoint of the ESN to compute two types of sensitivity: (i) parameter sensitivity, which is the gradient of a time-averaged cost functional with respect to physical or design parameters of the system, and (ii) initial condition sensitivity, which is the gradient of a cost functional of the final state with respect to the initial condition. Third, we propose the thermoacoustic echo state network (T-ESN), which embeds the physical knowledge in the network architecture for improved generalization. Fourth, we apply the framework to a variety of nonlinear thermoacoustic regimes of a prototypical system. We show that the T-ESN accurately infers the correct adjoint sensitivities of the acoustic energy with respect to the flame parameters and initial conditions. The results are robust to noisy data, from periodic, through quasiperiodic, to chaotic regimes. The inferred adjoint sensitivities are employed to suppress an instability via steepest descent. We show that a single network predicts the nonlinear bifurcations on unseen scenarios, which allows it to converge to the minimum of the acoustic energy. This work opens new possibilities for gradient-based data-driven design optimization. Published by the American Physical Society 2024

A novel FuseDecode Autoencoder for industrial visual inspection: Incremental anomaly detection improvement with gradual transition from unsupervised to mixed-supervision learning with reduced human effort

The industrial implementation of automated visual inspection leveraging deep learning is limited due to the labor-intensive labeling of datasets and the lack of datasets containing images of defects, which is especially the case in high-volume manufacturing with zero defect constraints. In this study, we present the FuseDecode Autoencoder (FuseDecode AE), a novel reconstruction-based anomaly detection model featuring incremental learning. Initially, the FuseDecode AE operates in an unsupervised manner on noisy data containing predominantly normal images and a small number of anomalous images. The predictions generated assist experts in distinguishing between normal and anomalous samples. Later, it adapts to weakly labeled datasets by retraining in a semi-supervised manner on normal data augmented with synthetic anomalies. As more real anomalous samples become available, the model further refines its capabilities through mixed-supervision learning on both normal and anomalous samples. Evaluation on a real industrial dataset of coating defects shows the effectiveness of the incremental learning approach. Furthermore, validation on the publicly accessible MVTec AD dataset demonstrates the FuseDecode AE's superiority over other state-of-the-art reconstruction-based models. These findings underscore its generalizability and effectiveness in automated visual inspection tasks, particularly in industrial settings.

A Self-Distilled Learning to Rank Model for Ad Hoc Retrieval

Learning to rank models are broadly applied in ad hoc retrieval for scoring and sorting documents based on their relevance to textual queries. The generalizability of the trained model in the learning to rank approach, however, can have an impact on the retrieval performance, particularly when data includes noise and outliers, or is incorrectly collected or measured. In this paper, we introduce a Self-Distilled Learning to Rank (SDLR) framework for ad hoc retrieval, and analyze its performance over a range of retrieval datasets and also in the presence of features’ noise. SDLR assigns a confidence weight to each training sample, aiming at reducing the impact of noisy and outlier data in the training process. The confidence weight is approximated based on the feature’s distributions derived from the values observed for the features of the documents labeled for a query in a listwise training sample. SDLR includes a distillation process that facilitates passing on the underlying patterns in assigning confidence weights from the teacher model to the student one. We empirically illustrate that SDLR outperforms state-of-the-art learning to rank models in ad hoc retrieval. We thoroughly investigate the SDLR performance in different settings including when no distillation strategy is applied; when different portion of data are used for training the teacher and the student models, and when both teacher and student models are trained over identical data. We show that SDLR is more effective when training data are split between a teacher and a student model. We also show that SDLR’s performance is robust when data features are noisy.

Cascaded redundant convolutional encoder-decoder network improved apnea detection performance using tracheal sounds in post anesthesia care unit patients.


Methods of detecting apnea based on acoustic features can be prone to misdiagnosed and missed diagnoses due to the influence of noise. The aim of this paper is to improve the performance of apnea detection algorithms in the Post Anesthesia Care Unit (PACU) using a denoising method that processes tracheal sounds without the need for separate background noise.
Approach:
Tracheal sound data from laboratory subjects was collected using a microphone. Record a segment of clinical background noise and clean tracheal sound data to synthesize the noisy tracheal sound data according to a specified signal-to-noise ratio. Extract the frequency-domain features of the tracheal sounds using the Short Time Fourier Transform (STFT) and input the Cascaded Redundant Convolutional Encoder-Decoder network (CR-CED) network for training. Patients' tracheal sound data collected in the PACU were then fed into the CR-CED network as test data and inversely transformed by STFT to obtain denoised tracheal sounds. The apnea detection algorithm was used to detect the tracheal sound after denoising.
Results:
Apnea events were correctly detected 207 times and normal respiratory events 11,305 times using tracheal sounds denoised by the CR-CED network. The sensitivity and specificity of apnea detection were 88% and 98.6%, respectively.
Significance:
The apnea detection results of tracheal sounds after CR-CED network denoising in the PACU are accurate and reliable. Tracheal sound can be denoised using this approach without separate background noise. It effectively improves the applicability of the tracheal sound denoising method in the medical environment while ensuring its correctness.&#xD.

Multi-Step Temperature Prognosis of Lithium-Ion Batteries for Real Electric Vehicles Based on a Novel Bidirectional Mamba Network and Sequence Adaptive Correlation

The battery systems of electric vehicles (EVs) are directly impacted by battery temperature in terms of thermal runaway and failure. However, uncertainty about thermal runaway, dynamic conditions, and a dearth of high-quality data sets make modeling and predicting nonlinear multiscale electrochemical systems challenging. In this work, a novel Mamba network architecture called BMPTtery (Bidirectional Mamba Predictive Battery Temperature Representation) is proposed to overcome these challenges. First, a two-step hybrid model of trajectory piecewise–polynomial regression and exponentially weighted moving average is created and used to an operational dataset of EVs in order to handle the problem of noisy and incomplete time-series data. Each time series is then individually labeled to learn the representation and adaptive correlation of the multivariate series to capture battery performance variations in complex dynamic operating environments. Next, a prediction method with multiple steps based on the bidirectional Mamba is suggested. When combined with a failure diagnosis approach, this scheme can accurately detect heat failures due to excessive temperature, rapid temperature rise, and significant temperature differences. The experimental results demonstrate that the technique can accurately detect battery failures on a dataset of real operational EVs and predict the battery temperature one minute ahead of time with an MRE of 0.273%.

Modeling and Hybrid Inversion of Mineral Deposits Using the Dipping Dike Model with Finite Depth Extent

The dipping dike model has shown to be a useful approximation for mineral deposits. To make this model more realistic, we include the thickness, which yields the depth to the bottom, as an additional parameter. The magnetic anomaly is obtained by combining the anomalies of two infinite dikes, so that the resulting expression is simpler than the classical prismatic models with polygonal cross section. We employ a Metropolis-Hasting (MH) algorithm coupled with the Levenberg-Marquardt (LM) method to invert magnetic profiles assuming a model of multiple dike-like sources. We use a few iterations of the LM method to improve the candidate solutions at the end of each random walk generated by MH. The following parameters are obtained: depth to the top, thickness, half-width, horizontal location of the top center, geological dip, in addition to two effective parameters that depend on the intensity of magnetization and the directions of the induced and remanent fields. For synthetic anomalies, both noise-free and noisy magnetic data are considered, with examples presented for each scenario. These examples highlight the discrepancy between models with finite and infinite sources. They also illustrate the higher accuracy of the hybrid MH-LM method over the pure MH approach. Moreover, two field examples related to mineral exploration have been considered: the Pima copper mine, United States, where the relative differences between the parameters obtained by our algorithm and those known from drilling are not higher than 10%, and a magnetic profile over iron ore deposits located in Laje, northeast Brazil, where the inverted parameters were useful for detailing previous studies.

Mango leaf disease diagnosis using Total Variation Filter Based Variational Mode Decomposition

Mango leaf diseases significantly threaten mango cultivation, impacting both yield and quality. Accurate and early diagnosis is essential for effectively managing and controlling these diseases. This study introduces a novel approach for diagnosing mango leaf diseases, leveraging Total Variation Filter-based Variational Mode Decomposition. The proposed method enhances the extraction of disease-specific features from leaf images by decomposing them into intrinsic mode functions while simultaneously reducing noise and preserving important edge information. Experimental results demonstrate that the proposed method effectively isolates relevant patterns associated with various mango leaf diseases, improving diagnostic accuracy compared to traditional methods. Deep learning models, DenseNet121 and VGG-19, are used for feature extraction from sub-band images, and extracted features are concatenated and fed to Random Forest for classification. Utilizing tenfold cross-validation, our model demonstrated enhanced classification accuracy (98.85 %), specificity (99.37 %), and sensitivity (98.0 %) in detecting diseases from Mango leaf images. Feature maps and Gradient-weighted Class Activation Mapping analysis was conducted to visualize and scrutinize the essential regions crucial for accurate predictions. Statistical analysis indicates that our proposed architecture outperforms pre-trained models and existing mango leaf disease detection methods. This diagnostic approach can be a rapid disease detection tool for imaging specialists utilizing leaf images. The robustness and efficiency of the presented work in handling complex and noisy image data make it a promising tool for automated agricultural disease diagnosis systems, facilitating timely and precise interventions in mango orchards.

Exploring noise-induced techniques to strengthen deep learning model resilience

AbstractIn artificial intelligence, combating overfitting and enhancing model generalization is crucial. This research explores innovative noise-induced regularization techniques, focusing on natural language processing tasks. Inspired by gradient noise and Dropout, this study investigates the interplay between controlled noise, model complexity, and overfitting prevention. Utilizing long short-term memory and bidirectional long short term memory architectures, this study examines the impact of noise-induced regularization on robustness to noisy input data. Through extensive experimentation, this study shows that introducing controlled noise improves model generalization, especially in language understanding. This contributes to the theoretical understanding of noise-induced regularization, advancing reliable and adaptable artificial intelligence systems for natural language processing.

Identification of potential biomarkers of papillary thyroid carcinoma.

This study aimed to identify candidate biomarkers for papillary thyroid carcinoma using an integrative analysis of bioinformatics and machine learning (ML). The PTC datasets GSE6004, GSE3467, and GSE33630 (species: Homo sapiens) were downloaded from NCBI and analyzed using the limma package to obtain DEGs. Once DEGs were identified, GO and KEGG enrichment analyses were performed as the first step in the bioinformatics process. Subsequently, a protein-protein interaction (PPI) network was constructed according to the common genes in bioinformatics and machine learning using STRING to elucidate the important genes involved in PTC pathogenesis. In machine learning, finding genes entails feature selection to identify the key genes that distinguish biological states. Hybrid feature selection will be used for this. In the second step, the original data sets were preprocessed to detect and correct missing and noisy data; after that, all data were merged. Following performing Linear and Discriminative Hybrid Feature Selection (LDHFS) on the processed dataset, machine learning algorithms such as Random Forest (RF), Naive Bayes (NB), and Support Vector Machines (SVM) are utilized. Bioinformatics and machine learning analyses indicate that the genes RXRG, CDH2, ETV5, QPCT, LRP4, FN1, and LPAR5 are integral to the progression of thyroid cancer. This study attained the highest accuracy utilizing the RF algorithm, achieving an accuracy rate of 94.62%, a Kappa value of 91.36%, and an AUC value of 96.13%. These results offer additional evidence and confirmation for the genetic alterations of these genes. These findings may accelerate the development of prospective therapeutic and diagnostic methods in future research. Bioinformatics and machine learning techniques identified the common genes "RXRG, CDH2, ETV5, QPCT, LRP4, FN1, and LPAR5" as PTC biomarkers, providing novel reference markers for the diagnosis and treatment of PTC patients. The model is anticipated to possess significant predictive value and assist in the early diagnosis and screening of clinical PTC. These insights enhance the field of PTC management and offer guidance for future research.

Unsupervised deep denoising for four-dimensional scanning transmission electron microscopy

By simultaneously achieving high spatial and angular sampling resolution, four dimensional scanning transmission electron microscopy (4D STEM) is enabling analysis techniques that provide great insight into the atomic structure of materials. Applying these techniques to scientifically and technologically significant beam-sensitive materials remains challenging because the low doses needed to minimise beam damage lead to noisy data. We demonstrate an unsupervised deep learning model that leverages the continuity and coupling between the probe position and the electron scattering distribution to denoise 4D STEM data. By restricting the network complexity it can learn the geometric flow present but not the noise. Through experimental and simulated case studies, we demonstrate that denoising as a preprocessing step enables 4D STEM analysis techniques to succeed at lower doses, broadening the range of materials that can be studied using these powerful structure characterization techniques.

Emergency Detection in Smart Homes Using Inactivity Score for Handling Uncertain Sensor Data

In an aging society, the need for efficient emergency detection systems in smart homes is becoming increasingly important. For elderly people living alone, technical solutions for detecting emergencies are essential to receiving help quickly when needed. Numerous solutions already exist based on wearable or ambient sensors. However, existing methods for emergency detection typically assume that sensor data are error-free and contain no false positives, which cannot always be guaranteed in practice. Therefore, we present a novel method for detecting emergencies in private households that detects unusually long inactivity periods and can process erroneous or uncertain activity information. We introduce the Inactivity Score, which provides a probabilistic weighting of inactivity periods based on the reliability of sensor measurements. By analyzing historical Inactivity Scores, anomalies that potentially represent an emergency can be identified. The proposed method is compared with four related approaches on seven different datasets. Our method surpasses existing approaches when considering the number of false positives and the mean time to detect emergencies. It achieves an average detection time of approximately 05:23:28 h with only 0.09 false alarms per day under noise-free conditions. Moreover, unlike related approaches, the proposed method remains effective with noisy data.

II-LA-KM: Improved Initialization of a Learning-Augmented Clustering Algorithm for Effective Rock Discontinuity Grouping

Rock mass discontinuities are an excellent information set for reflecting the geometric, spatial, and physical properties of the rock mass. Using clustering algorithms to analyze them is a significant way to select advantageous orientations of structural surfaces and provide a scientific theoretical basis for other rock mass engineering research. Traditional clustering algorithms often suffer from sensitivity to initialization and lack practical applicability, as discontinuity data are typically rough, low-precision, and unlabeled. Confronting these challenges, II-LA-KM, a learning-augmented clustering algorithm with improved initialization for rock discontinuity grouping, is proposed. Our method begins with heuristically selecting initial centers to ensure they are well-separated. Then, optimal transport is used to adjust these centers, minimizing the transport cost between them and other points. To enhance fault tolerance, a learning-augmented algorithm is integrated that iteratively reduces clustering costs, refining the initial results toward optimal clustering. Extensive experiments on a simulated artificial dataset and a real dataset from Woxi, Hunan, China, featuring both orientational and non-orientational attributes, demonstrate the effectiveness of II-LA-KM. The algorithm achieves a 97.5% accuracy on the artificial dataset and successfully differentiates between overlapping groups. Its performance is even more pronounced on the real dataset, underscoring its robustness for handling complex and noisy data. These strengths make our approach highly beneficial for practical rock discontinuity grouping applications.

AnFiS-MoH: Systematic exploration of hybrid ANFIS frameworks via metaheuristic optimization hybridization with evolutionary and swarm-based algorithms

The adaptive neuro-fuzzy inference system (ANFIS) has shown promising performance in modeling nonlinear problems, leveraging the strengths of both neural networks and fuzzy inference systems. However, as the problem scale increases, the growing number of tunable parameters in ANFIS can make it challenging to optimize via traditional gradient-based methods alone. This study introduces AnFiS-MoH, a novel framework that synergistically integrates ANFIS with metaheuristic optimization algorithms to address these challenges. By leveraging the global search capabilities of metaheuristics such as ant colony optimization (ACO), particle swarm optimization (PSO), genetic algorithm (GA), and simulated annealing (SA), ANFIS-MOH enhances the parameter tuning process of ANFIS models. We evaluate ANFIS-MOH on benchmark datasets including Boston Housing and Wine Quality, demonstrating significant improvements in prediction accuracy and generalization compared to traditional ANFIS and neural network approaches. The proposed framework achieves up to 20% reduction in Mean Squared Error and 15% increase in R2 scores, particularly excelling in handling high-dimensional, noisy data. This work contributes to the field of hybrid intelligent systems by introducing effective ways to combine the strengths of ANFIS with powerful metaheuristic optimization algorithms. The findings suggest that such hybrid approaches can be effective in tackling challenging nonlinear modeling problems. Our code is available at https://github.com/AmbitYuki/Metaheuristic-Adaptive-ANFIS.

Three-dimensional visualization of large-area, nanoscale topography measurements

High-resolution metrology is a critical area of development for nanoscale manufacturing, especially as it affects production throughput and fabrication quality. Atomic force microscopy (AFM) is one of the most popular tools for nanometrology, and high-resolution AFM often requires a significant time commitment and produces datasets of several million points. It is therefore critical for the development of data processing techniques to keep pace with the requirements of analyzing this type of data, and for these techniques to be portable as miniaturization in AFM is becoming more common. This work presents a data fitting algorithm designed for reducing the parameters of large-area data sets which utilizes well-established spline fitting techniques. In this paper we show that basis-spline fitting can be used to accurately represent large AFM data sets, including data sets with noisy data and sharp features, while achieving at least 90% parameter reduction in all test cases.

MTC-NET: A Multi-Channel Independent Anomaly Detection Method for Network Traffic.

In recent years, deep learning-based approaches, particularly those leveraging the Transformer architecture, have garnered widespread attention for network traffic anomaly detection. However, when dealing with noisy data sets, directly inputting network traffic sequences into Transformer networks often significantly degrades detection performance due to interference and noise across dimensions. In this paper, we propose a novel multi-channel network traffic anomaly detection model, MTC-Net, which reduces computational complexity and enhances the model's ability to capture long-distance dependencies. This is achieved by decomposing network traffic sequences into multiple unidimensional time sequences and introducing a patch-based strategy that enables each sub-sequence to retain local semantic information. A backbone network combining Transformer and CNN is employed to capture complex patterns, with information from all channels being fused at the final classification header in order to achieve modelling and detection of complex network traffic patterns. The experimental results demonstrate that MTC-Net outperforms existing state-of-the-art methods in several evaluation metrics, including accuracy, precision, recall, and F1 score, on four publicly available data sets: KDD Cup 99, NSL-KDD, UNSW-NB15, and CIC-IDS2017.

Pose‐to‐Motion: Cross‐Domain Motion Retargeting with Pose Prior

AbstractCreating plausible motions for a diverse range of characters is a long‐standing goal in computer graphics. Current learning‐based motion synthesis methods rely on large‐scale motion datasets, which are often difficult if not impossible to acquire. On the other hand, pose data is more accessible, since static posed characters are easier to create and can even be extracted from images using recent advancements in computer vision. In this paper, we tap into this alternative data source and introduce a neural motion synthesis approach through retargeting, which generates plausible motion of various characters that only have pose data by transferring motion from one single existing motion capture dataset of another drastically different characters. Our experiments show that our method effectively combines the motion features of the source character with the pose features of the target character, and performs robustly with small or noisy pose data sets, ranging from a few artist‐created poses to noisy poses estimated directly from images. Additionally, a conducted user study indicated that a majority of participants found our retargeted motion to be more enjoyable to watch, more lifelike in appearance, and exhibiting fewer artifacts. Our code and dataset can be accessed here.