Unsupervised Deep Learning Research Articles

Nondestructive detection of laser-cladding coating interface defects: a deep learning-enhanced laser ultrasonics approach

ABSTRACT Laser-cladding technology is now widely used in repairing mechanical parts and creating functional coatings. However, interface defects are commonly found in laser-cladding coatings, significantly impacting the fatigue and mechanical properties of these coatings. Therefore, it is necessary to detect interface defects in laser-cladding coatings. In this study, we utilised laser ultrasonics (LU) method to test interface defects non-destructively. However, the signal-to-noise ratio (SNR) of LU signals was poor. To address this issue, an unsupervised deep learning method was used to enhance the SNR. The results showed that the noises were suppressed by the deep learning method. Based on the deep learning-enhanced LU method, surface acoustic waves were used to present B-scan and C-scan images. The artificial defects with a 0.5 mm diameter were detected. Consequently, the deep learning-enhanced LU method proves to be a viable approach for the contactless, online, and non-destructive testing of interface defects in laser-cladding coatings.

Energy landscapes of homopolymeric RNAs revealed by deep unsupervised learning

Conformational dynamics of RNA plays important roles in a variety of cellular functions such as transcriptional regulation, catalysis, scaffolding, and sensing. Recently, RNAs with low-complexity sequences have been shown to phase separate and form condensate phases similar to lowcomplexity protein domains. The affinity for phase separation and the material characteristics of RNA condensates are strongly dependent on sequence composition and patterning. We hypothesize that differences in the affinities for RNA phase separation can be uncovered by studying sequence-dependent conformational dynamics of single RNA chains. To this end, we have employed atomistic simulations and deep dimensionality reduction techniques to map temperature-dependent conformational free energy landscapes for 20 base-long homopolymeric RNA sequences: poly(U), poly(G), poly(C), and poly(A). The energy landscapes of homopolymeric RNAs reveal a plethora of metastable states with qualitatively different populations stemming from differences in base chemistry. Through detailed analysis of base, phosphate, and sugar interactions, we show that experimentally observed temperature-driven shifts in metastable state populations align with experiments on RNA phase transitions. Specifically, we find that the thermodynamics of unfolding of homopolymeric RNA follows the poly(G) > poly(A) > poly(C) > poly(U) order of stability, mirroring the propensity of RNA to form condensates. To conclude, this work shows that at least for homopolymeric RNA sequences the single-chain conformational dynamics contains sufficient information for predicting and quantifying condensate forming affinities of RNAs. Thus, we anticipate that atomically detailed studies of temeprature -dependent energy landscapes of RNAs will be a useful guide for understanding the propensity of various RNA molecules to form condensates.

Deep Image Prior-Based PET Reconstruction From Partial Data

In this paper, we propose an unsupervised deep learning method for positron emission tomography reconstruction (PET) from incomplete data. This method utilizes the so-called deep image prior (DIP) as an untrained deep convolutional neural network (CNN) to generate object reconstructions. The main idea is to re-parameterize the image reconstruction problem as a neural network optimization problem. We show that the proposed method effectively addresses the incomplete data reconstruction problem, which otherwise degrades the image resolution and quality of standard reconstruction algorithms. Meanwhile, the proposed method does not require any pre-training procedures, i.e., it is not biased toward any particular dataset. Hence, it has the potential to be used in clinical situations, where training data would be infeasible or prohibitively expensive. The performance of the proposed approach is evaluated with noisy synthetic data based on shepp-logan and brainweb phantoms, and clinical naive rat data. In addition, robustness studies of the approach with respect to regularization parameters are also carried out. We showcase that the proposed method considerably outperforms the state-of-the-art methods, leading to flexible reconstruction from incomplete PET data.

Attention non-negative spectral clustering

Deep spectral clustering is an advanced unsupervised deep learning method with widespread applications. However, when applied to large-scale datasets, it may confront problems such as an unstable training process, limited scalability, and diminished interpretability. To mitigate these limitations, a novel unsupervised deep learning algorithm named Attention Non-negative Spectral Clustering (ANSC) is proposed. Specifically, a network architecture called the Spectral Attention Network (SAN) is designed to serve as the backbone network for ANSC. This design aims to improve the stability of the training process and the model’s scalability. This attention-based network architecture includes a feature extraction network, an eigenvector mapping network, and an orthogonalization module. Furthermore, to enhance the interpretability of ANSC, the objective function incorporates a cluster indicator matrix by introducing non-negative constraints. Additionally, the ℓ2,1-norm is employed as regularization to approximate the solution of the objective function. An analysis of a lower bound for the network size of ANSC is conducted based on the VC dimension. Extensive experiments validate that ANSC outperforms state-of-the-art clustering methods in terms of clustering accuracy, normalized mutual information, and adjusted rand index.

Enhancing Intrusion Detection through Unsupervised Deep Learning Models

The Internet of Things (IoT) has experienced significant growth since its inception, representing a groundbreaking technological advancement. In essence, IoT involves the seamless integration of devices and data to automate and centralize various processes. This transformative technology is revolutionizing business operations and reshaping society as a whole. As IoT continues to evolve, the importance of detecting vulnerabilities and weaknesses becomes paramount in order to thwart unauthorized access to critical resources and business functions, which could potentially render the entire system unavailable. One prevalent threat in this context is Denial of Service (DoS) and Distributed DoS attacks. In this project, propose an innovative architecture known as Protocol Based Deep Intrusion Detection (PB-DID). To create our dataset, we gathered packets from IoT traffic and compared features from two well-known datasets: UNSWNB15 and Bot-IoT. We focused on flow and Transmission Control Protocol (TCP) characteristics. Our primary objective was to accurately classify network traffic into three categories: non-anomalous, DoS, and DDoS, while addressing issues like data imbalance and overfitting. Utilizing deep learning (DL) techniques, we achieved an impressive classification accuracy of 96.3%. This level of accuracy demonstrates the potential of our PB-DID architecture in effectively identifying and mitigating intrusion attempts in the context of IoT, thereby enhancing the security and reliability of IoT systems. Keywords—: Intrusion detection in IoT, Deep learning for intrusion detection, DoS detection, DDoS detection.

Environment‐Modulated Glacial Seismicity Near Dålk Glacier in East Antarctica Revealed by Deep Clustering

AbstractOver the past decades, seismic monitoring has been increasingly used to track glacial activities associated with ice loss. Many seismological studies focus on West Antarctica, whereas glacial seismicity in East Antarctica is much less studied. Here, we apply unsupervised deep learning to a dense nodal seismic array near Dålk Glacier, East Antarctica, operating from 6 December 2019 to 2 January 2020. An autoencoder is used to automatically extract event features, which are then input into a Gaussian mixture model for clustering. We divide the data into 50 clusters and merge them according to their temporal and spectral characteristics. The results reveal five main types of seismic signals: two groups with monochromatic and high frequencies, two groups with broadband frequency and short duration, and a group with mainly low frequency and long duration. By comparing the environmental conditions (wind, temperature and tides), we infer that the two monochromatic groups are wind‐induced vibrations of the near‐station flag markers and topography; the two broadband groups are likely thermal contractions on the blue ice surface and stick‐slip events at the ice base; and the low‐frequency events are water‐filled basal crevassing and iceberg calving. In particular, we observe one type of low‐frequency event preceded by high‐frequency onset, which is likely basal crevassing near the grounding line of Dålk Glacier and predominantly occurred during rising tides. Our findings show that deep clustering is effective in identifying a wide range of glacial seismic events and can contribute to the rapid growth of passive glacier seismic monitoring.

Unsupervised Denoising and Super-Resolution of Vascular Flow Data by Physics-Informed Machine Learning.

We present an unsupervised deep learning method to perform flow denoising and super-resolution without high-resolution labels. We demonstrate the ability of a single model to reconstruct three-dimensional stenosis and aneurysm flows, with varying geometries, orientations, and boundary conditions. Ground truth data was generated using computational fluid dynamics, and then corrupted with multiplicative Gaussian noise. Auto-encoders were used to compress the representations of the flow domain geometry and the (possibly noisy and low-resolution) flow field. These representations were used to condition a physics-informed neural network. A physics-based loss was implemented to train the model to recover lost information from the noisy input by transforming the flow to a solution of the Navier-Stokes equations. Our experiments achieved mean squared errors in the true flow reconstruction of O(1.0 × 10-4), and root mean squared residuals of O(1.0 × 10-2) for the momentum and continuity equations. Our method yielded correlation coefficients of 0.971 for the hidden pressure field and 0.82 for the derived wall shear stress field. By performing point-wise predictions of the flow, the model was able to robustly denoise and super-resolve the field to 20× the input resolution.

DL-EDOF: Novel Multi-Focus Image Data Set and Deep Learning-Based Approach for More Accurate and Specimen-Free Extended Depth of Focus.

Depth of focus (DOF) is defined as the axial range in which the specimen stage moves without losing focus while the imaging apparatus remains stable. It may not be possible to capture an image that includes the entire specimen in focus due to the narrow DOF in microscopic systems. Extended depth of focus (EDOF) is used to overcome this limitation in microscopic systems. Although the researchers have developed so many EDOF microscope approaches, this research field still has some crucial shortcomings such as high computational costs, complexity and execution time, requiring additional equipment, low precise characterization of curves, and edges in images, varying performance depending on the specimen and microscope, using only gray levels of input images to acquire the pixel's focus values. In order to minimize these shortcomings and comprehensively analyze the performance of EDOF approaches, a novel multi-focus image data set is generated, and a deep learning-based EDOF microscope approach is proposed in this study. When compared with the state-of-art EDOF approaches, our study provides various crucial contributions such as the first EDOF approach based on unsupervised deep learning, providing more accurate and specimen-free EDOF, generating a novel multi-focus image data, not requiring any pre- or post-processing technique and acquiring the pixel's focus degrees using deep features. In order to evaluate the effectiveness of the suggested approach, 20 different EDOF approaches are applied to a multi-focus image data set containing 9 image collections (4 synthetic and 5 microscope image collections) in total. Performance analysis metrics with and without requiring a reference image are preferred to identify which EDOF microscope approach can extract more essential details from the multi-focus images for the synthetic and microscope image collections, which are Root Mean Square Error (RMSE), Peak Signal Noise Ratio (PSNR), Universal Quality Index (UQI), Correlation Coefficient (CC), Perception-based Image Quality Evaluator (PIQE), Blind/Referenceless Image Spatial Quality Evaluator (BRISQUE), Extension of Universal Quality Index for N Images (UQIN), and Naturalness Image Quality Evaluator (NIQE). Objective and subjective analysis of this study demonstrates that unsupervised deep learning model is more efficient to transmit crucial details from multi-focus images. Moreover, the suggested EDOF microscope approach with highest PSNR, UQI, CC, UQIN and lowest RMSE, PIQE, BRISQUE, NIQE produces higher performance than the state-of-art approaches.

Dual-Channel in Spatial-Frequency Domain CycleGAN for perceptual enhancement of transcranial cortical vascular structure and function

Observing cortical vascular structures and functions using laser speckle contrast imaging (LSCI) at high resolution plays a crucial role in understanding cerebral pathologies. Usually, open-skull window techniques have been applied to reduce scattering of skull and enhance image quality. However, craniotomy surgeries inevitably induce inflammation, which may obstruct observations in certain scenarios. In contrast, image enhancement algorithms provide popular tools for improving the signal-to-noise ratio (SNR) of LSCI. The current methods were less than satisfactory through intact skulls because the transcranial cortical images were of poor quality. Moreover, existing algorithms do not guarantee the accuracy of dynamic blood flow mappings. In this study, we develop an unsupervised deep learning method, named Dual-Channel in Spatial-Frequency Domain CycleGAN (SF-CycleGAN), to enhance the perceptual quality of cortical blood flow imaging by LSCI. SF-CycleGAN enabled convenient, non-invasive, and effective cortical vascular structure observation and accurate dynamic blood flow mappings without craniotomy surgeries to visualize biodynamics in an undisturbed biological environment. Our experimental results showed that SF-CycleGAN achieved a SNR at least 4.13 dB higher than that of other unsupervised methods, imaged the complete vascular morphology, and enabled the functional observation of small cortical vessels. Additionally, the proposed method showed remarkable robustness and could be generalized to various imaging configurations and image modalities, including fluorescence images, without retraining.

Abstract 2310: Integrating real-world histopathological and clinicogenomic data from 1799 lung cancer patients by applying unsupervised deep learning

Abstract Hematoxylin and eosin (H&E) stained histologic sections contain invaluable information that remains largely untapped because of its complexity. To this end, AI applications employing deep learning (DL) can facilitate the translation of image data to enable human interpretation and yield novel oncological insights that would have otherwise remained imperceptible. DL-based methodologies are multimodal, capable of integrating imaging with clinicogenomic data to furnish a more holistic perspective and affording more accurate predictions in oncology. Here, we developed an unsupervised DL workflow to analyze 1,799 H&E images of lung cancer (NSCLC n = 951; SCLC n = 50; others n = 798) incorporating comprehensive patient-level clinical data (electronic health records [ConcertAI]) integrated with genomics (WES and RNA-seq [Caris Labs]). There are three steps in our approach: (1) image preprocessing and filtering, yielding > 30 million image patches; (2) utilizing pretrained SimCLR models from 57 public oncology histopathology datasets with ResNet-18 as a backbone structure to extract 512-dimensional-feature vectors for each patch; (3) using three unsupervised clustering methods (kmeans, DBSCAN, Leiden clustering) to cluster patches and selected Leiden clustering. We identified 635 primary imaging clusters using an elbow method and generated an image feature matrix by calculating correlations between each patch and cluster centroids; these were aggregated and mapped back to source slides. In this proof-of-concept, distinct image feature patterns characterized SCLC and NSCLC samples. For SCLC, one of the salient features was the presence of hemorrhage, which may be associated with higher rates of fine-needle aspiration biopsy procedure for SCLC compared with NSCLC which was confirmed in the EHR data (p = 0.032). Derived morphological clusters were correlated with tumor-immune genomic features (Tumor Mutational Burden [TMB], Immunologic Constant of Rejection [ICR], and Miracle scores1) serving as predictors of response to immune-checkpoint inhibitor therapy. By applying linear models, we detected 11, 96 and 249 significantly associated imaging clusters, respectively, highly enriched with immune cells e.g., plasma cells, macrophages, lymphocytes, and supporting an infiltrated and inflamed tumor-immune microenvironment. In summary, a multimodal, unsupervised deep learning workflow combining H&E imaging with clinicogenomic data was developed to identify histologic feature clusters associated with well-established tumor-immune genomic signatures of NSCLC immune infiltration and molecular phenotypes. These studies demonstrate enormous potential to yield histopathological and translational insights in NSCLC and SCLC that can empower clinicians to make better therapeutic response predictions. Citation Format: Si Wu, Yujie Zhao, Hugo Luo, Kevin Kolahi, Thanh Bui, Xu Shi, Aditee Shrotre, Alexander Liede, Xi Zhao, Josue Samayoa, Weilong Zhao. Integrating real-world histopathological and clinicogenomic data from 1799 lung cancer patients by applying unsupervised deep learning [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2024; Part 1 (Regular Abstracts); 2024 Apr 5-10; San Diego, CA. Philadelphia (PA): AACR; Cancer Res 2024;84(6_Suppl):Abstract nr 2310.

An Efficient Convolutional Denoising Autoencoder-Based BDS NLOS Detection Method in Urban Forest Environments.

The BeiDou Navigation Satellite System (BDS) provides real-time absolute location services to users around the world and plays a key role in the rapidly evolving field of autonomous driving. In complex urban environments, the positioning accuracy of BDS often suffers from large deviations due to non-line-of-sight (NLOS) signals. Deep learning (DL) methods have shown strong capabilities in detecting complex and variable NLOS signals. However, these methods still suffer from the following limitations. On the one hand, supervised learning methods require labeled samples for learning, which inevitably encounters the bottleneck of difficulty in constructing databases with a large number of labels. On the other hand, the collected data tend to have varying degrees of noise, leading to low accuracy and poor generalization performance of the detection model, especially when the environment around the receiver changes. In this article, we propose a novel deep neural architecture named convolutional denoising autoencoder network (CDAENet) to detect NLOS in urban forest environments. Specifically, we first design a denoising autoencoder based on unsupervised DL to reduce the long time series signal dimension and extract the deep features of the data. Meanwhile, denoising autoencoders improve the model's robustness in identifying noisy data by introducing a certain amount of noise into the input data. Then, an MLP algorithm is used to identify the non-linearity of the BDS signal. Finally, the performance of the proposed CDAENet model is validated on a real urban forest dataset. The experimental results show that the satellite detection accuracy of our proposed algorithm is more than 95%, which is about an 8% improvement over existing machine-learning-based methods and about 3% improvement over deep-learning-based approaches.

Open-top Bessel beam two-photon light sheet microscopy for three-dimensional pathology

Nondestructive pathology based on three-dimensional (3D) optical microscopy holds promise as a complement to traditional destructive hematoxylin and eosin (H&E) stained slide-based pathology by providing cellular information in high throughput manner. However, conventional techniques provided superficial information only due to shallow imaging depths. Herein, we developed open-top two-photon light sheet microscopy (OT-TP-LSM) for intraoperative 3D pathology. An extended depth of field two-photon excitation light sheet was generated by scanning a nondiffractive Bessel beam, and selective planar imaging was conducted with cameras at 400 frames/s max during the lateral translation of tissue specimens. Intrinsic second harmonic generation was collected for additional extracellular matrix (ECM) visualization. OT-TP-LSM was tested in various human cancer specimens including skin, pancreas, and prostate. High imaging depths were achieved owing to long excitation wavelengths and long wavelength fluorophores. 3D visualization of both cells and ECM enhanced the ability of cancer detection. Furthermore, an unsupervised deep learning network was employed for the style transfer of OT-TP-LSM images to virtual H&E images. The virtual H&E images exhibited comparable histological characteristics to real ones. OT-TP-LSM may have the potential for histopathological examination in surgical and biopsy applications by rapidly providing 3D information.

Open-top Bessel beam two-photon light sheet microscopy for three-dimensional pathology.

Nondestructive pathology based on three-dimensional (3D) optical microscopy holds promise as a complement to traditional destructive hematoxylin and eosin (H&E) stained slide-based pathology by providing cellular information in high throughput manner. However, conventional techniques provided superficial information only due to shallow imaging depths. Herein, we developed open-top two-photon light sheet microscopy (OT-TP-LSM) for intraoperative 3D pathology. An extended depth of field two-photon excitation light sheet was generated by scanning a nondiffractive Bessel beam, and selective planar imaging was conducted with cameras at 400 frames/s max during the lateral translation of tissue specimens. Intrinsic second harmonic generation was collected for additional extracellular matrix (ECM) visualization. OT-TP-LSM was tested in various human cancer specimens including skin, pancreas, and prostate. High imaging depths were achieved owing to long excitation wavelengths and long wavelength fluorophores. 3D visualization of both cells and ECM enhanced the ability of cancer detection. Furthermore, an unsupervised deep learning network was employed for the style transfer of OT-TP-LSM images to virtual H&E images. The virtual H&E images exhibited comparable histological characteristics to real ones. OT-TP-LSM may have the potential for histopathological examination in surgical and biopsy applications by rapidly providing 3D information.

A Prescriptive Model for Failure Analysis in Ship Machinery Monitoring Using Generative Adversarial Networks

In recent years, advanced methods and smart solutions have been investigated for the safe, secure, and environmentally friendly operation of ships. Since data acquisition capabilities have improved, data processing has become of great importance for ship operators. In this study, we introduce a novel approach to ship machinery monitoring, employing generative adversarial networks (GANs) augmented with failure mode and effect analysis (FMEA), to address a spectrum of failure modes in diesel generators. GANs are emerging unsupervised deep learning models known for their ability to generate realistic samples that are used to amplify a number of failures within training datasets. Our model specifically targets critical failure modes, such as mechanical wear and tear on turbochargers and fuel injection system failures, which can have environmental effects, providing a comprehensive framework for anomaly detection. By integrating FMEA into our GAN model, we do not stop at detecting these failures; we also enable timely interventions and improvements in operational efficiency in the maritime industry. This methodology not only boosts the reliability of diesel generators, but also sets a precedent for prescriptive maintenance approaches in the maritime industry. The model was demonstrated with real-time data, including 33 features, gathered from a diesel generator installed on a 310,000 DWT oil tanker. The developed algorithm provides high-accuracy results, achieving 83.13% accuracy. The final model demonstrates a precision score of 36.91%, a recall score of 83.47%, and an F1 score of 51.18%. The model strikes a balance between precision and recall in order to eliminate operational drift and enables potential early action in identified positive cases. This study contributes to managing operational excellence in tanker ship fleets. Furthermore, this study could be expanded to enhance the current functionalities of engine health management software products.

GSLCDA: An Unsupervised Deep Graph Structure Learning Method for Predicting CircRNA-Disease Association.

Growing studies reveal that Circular RNAs (circRNAs) are broadly engaged in physiological processes of cell proliferation, differentiation, aging, apoptosis, and are closely associated with the pathogenesis of numerous diseases. Clarification of the correlation among diseases and circRNAs is of great clinical importance to provide new therapeutic strategies for complex diseases. However, previous circRNA-disease association prediction methods rely excessively on the graph network, and the model performance is dramatically reduced when noisy connections occur in the graph structure. To address this problem, this paper proposes an unsupervised deep graph structure learning method GSLCDA to predict potential CDAs. Concretely, we first integrate circRNA and disease multi-source data to constitute the CDA heterogeneous network. Then the network topology is learned using the graph structure, and the original graph is enhanced in an unsupervised manner by maximize the inter information of the learned and original graphs to uncover their essential features. Finally, graph space sensitive k-nearest neighbor (KNN) algorithm is employed to search for latent CDAs. In the benchmark dataset, GSLCDA obtained 92.67% accuracy with 0.9279 AUC. GSLCDA also exhibits exceptional performance on independent datasets. Furthermore, 14, 12 and 14 of the top 16 circRNAs with the most points GSLCDA prediction scores were confirmed in the relevant literature in the breast cancer, colorectal cancer and lung cancer case studies, respectively. Such results demonstrated that GSLCDA can validly reveal underlying CDA and offer new perspectives for the diagnosis and therapy of complex human diseases.

Semi-Supervised and Unsupervised Deep Visual Learning: A Survey.

State-of-the-art deep learning models are often trained with a large amount of costly labeled training data. However, requiring exhaustive manual annotations may degrade the model's generalizability in the limited-label regime.Semi-supervised learning and unsupervised learning offer promising paradigms to learn from an abundance of unlabeled visual data. Recent progress in these paradigms has indicated the strong benefits of leveraging unlabeled data to improve model generalization and provide better model initialization. In this survey, we review the recent advanced deep learning algorithms on semi-supervised learning (SSL) and unsupervised learning (UL) for visual recognition from a unified perspective. To offer a holistic understanding of the state-of-the-art in these areas, we propose a unified taxonomy. We categorize existing representative SSL and UL with comprehensive and insightful analysis to highlight their design rationales in different learning scenarios and applications in different computer vision tasks. Lastly, we discuss the emerging trends and open challenges in SSL and UL to shed light on future critical research directions.

Unsupervised deep learning approach for structural anomaly detection using probabilistic features

Civil structures may deteriorate during their service life due to degradation or damage imposed by natural hazards such as earthquakes, wind, and impact. Structural performance anomaly detection is essential to provide an early warning of structural degradation limit states in order to prevent potential catastrophic failure. Data-driven machine learning approaches have been widely used for this, due to their capability in capturing features sensitive to damage-induced anomalies from structural health monitoring (SHM) data, assuming that such data are available. Although machine learning models have been used, many are challenged by the vast operational and environmental variability that can corrupt SHM data and by (typically) strongly correlated information from different sensors in the SHM data. This paper proposes an unsupervised deep learning approach for the detection of structural anomaly based on a deep convolutional variational autoencoder (DCVAE) for feature extraction coupled with support vector data description (SVDD) for anomaly detection. The proposed DCVAE-SVDD method has several appealing strengths. First, the variational latent encoding is used to capture the features of monitoring data through a probability distribution. The integration of the Kullback–Leibler divergence in the loss function provides accurate estimation of the probability distributions. Second, the DCVAE designed with convolutional and deconvolutional operations utilizes the correlation among multisensor data to avoid loss of correlation features and achieve better performance in feature extraction. Third, the SVDD is utilized to create a minimum-volume hypersphere that contains the anomaly-sensitive statistical features of the state. The hypersphere accurately separates anomaly-sensitive statistical features of reference states of structure from the anomalous ones. A computational frame model and a laboratory grandstand model are used to evaluate the performance of the proposed method for detecting structural anomaly. The results demonstrate the superiority of the proposed DCVAE-SVDD in detection accuracy over the other commonly used structural anomaly detection methods (deep autoencoder combined with SVDD autoregressive model with one-class support vector machine, and principal component analysis).

An artificial intelligence platform for automated PFAS subgroup classification: A discovery tool for PFAS screening

Since structural analyses and toxicity assessments have not been able to keep up with the discovery of unknown per- and polyfluoroalkyl substances (PFAS), there is an urgent need for effective categorization and grouping of PFAS. In this study, we presented PFAS-Atlas, an artificial intelligence-based platform containing a rule-based automatic classification system and a machine learning-based grouping model. Compared with previously developed classification software, the platform's classification system follows the latest Organization for Economic Co-operation and Development (OECD) definition of PFAS and reduces the number of uncategorized PFAS. In addition, the platform incorporates deep unsupervised learning models to visualize the chemical space of PFAS by clustering similar structures and linking related classes. Through real-world use cases, we demonstrate that PFAS-Atlas can rapidly screen for relationships between chemical structure and persistence, bioaccumulation, or toxicity data for PFAS. The platform can also guide the planning of the PFAS testing strategy by showing which PFAS classes urgently require further attention. Ultimately, the release of PFAS-Atlas will benefit both the PFAS research and regulation communities.

Multimodal learning with only image data: A deep unsupervised model for street view image retrieval by fusing visual and scene text features of images

AbstractAs one of the classic tasks in information retrieval, the core of image retrieval is to identify the images sharing similar features with a query image, aiming to enable users to find the required information from a large number of images conveniently. Street view image retrieval, in particular, finds extensive applications in many fields, such as improvements to navigation and mapping services, formulation of urban development planning scheme, and analysis of historical evolution of buildings. However, the intricate foreground and background details in street view images, coupled with a lack of attribute annotations, render it among the most challenging issues in practical applications. Current image retrieval research mainly uses the visual model that is completely dependent on the image visual features, and the multimodal learning model that necessitates additional data sources (e.g., annotated text). Yet, creating annotated datasets is expensive, and street view images, which contain a large amount of scene texts themselves, are often unannotated. Therefore, this paper proposes a deep unsupervised learning algorithm that combines visual and text features from image data for improving the accuracy of street view image retrieval. Specifically, we employ text detection algorithms to identify scene text, utilize the Pyramidal Histogram of Characters encoding predictor model to extract text information from images, deploy deep convolutional neural networks for visual feature extraction, and incorporate a contrastive learning module for image retrieval. Upon testing across three street view image datasets, the results demonstrate that our model holds certain advantages over the state‐of‐the‐art multimodal models pre‐trained on extensive datasets, characterized by fewer parameters and lower floating point operations. Code and data are available at https://github.com/nwuSY/svtRetrieval.

Anomaly detection in automated fibre placement: learning with data limitations

Introduction: Conventional defect detection systems in Automated Fibre Placement (AFP) typically rely on end-to-end supervised learning, necessitating a substantial number of labelled defective samples for effective training. However, the scarcity of such labelled data poses a challenge.Methods: To overcome this limitation, we present a comprehensive framework for defect detection and localization in Automated Fibre Placement. Our approach combines unsupervised deep learning and classical computer vision algorithms, eliminating the need for labelled data or manufacturing defect samples. It efficiently detects various surface issues while requiring fewer images of composite parts for training. Our framework employs an innovative sample extraction method leveraging AFP’s inherent symmetry to expand the dataset. By inputting a depth map of the fibre layup surface, we extract local samples aligned with each composite strip (tow).Results: These samples are processed through an autoencoder, trained on normal samples for precise reconstructions, highlighting anomalies through reconstruction errors. Aggregated values form an anomaly map for insightful visualization. The framework employs blob detection on this map to locate manufacturing defects.Discussion: The experimental findings reveal that despite training the autoencoder with a limited number of images, our proposed method exhibits satisfactory detection accuracy and accurately identifies defect locations. Our framework demonstrates comparable performance to existing methods, while also offering the advantage of detecting all types of anomalies without relying on an extensive labelled dataset of defects.