Search

Temporal point process is widely used for sequential data modeling. In this paper, we focus on the problem of modeling sequential event propagation in graph, such as retweeting by social network users, news transmitting between websites, etc. Given a collection of event propagation sequences, conventional point process model consider only the event history, i.e. embed event history into a vector, not the latent graph structure. We propose a Graph Biased Temporal Point Process (GBTPP) leveraging the structural information from graph representation learning, where the direct influence between nodes and indirect influence from event history is modeled respectively. Moreover, the learned node embedding vector is also integrated into the embedded event history as side information. Experiments on a synthetic dataset and two real-world datasets show the efficacy of our model compared to conventional methods and state-of-the-art., Comment: 9 pages, 6 figures, 2 tables

The problem of representative selection amounts to sampling few informative exemplars from large datasets. Existing approaches to data selection often fall short of simultaneously handling non-linear data structures, sampling concise and non-redundant subsets, rejecting outliers, and yielding interpretable outcomes. This paper presents a novel representative selection approach, dubbed MOSAIC, for drawing descriptive sketches of arbitrary manifold structures. Resting upon a novel quadratic formulation, MOSAIC advances a multi-criteria selection approach that maximizes the global representation power of the sampled subset, ensures novelty of the samples by minimizing redundancy, and rejects disruptive information by effectively detecting outliers. Theoretical analyses shed light on geometrical characterization of the obtained sketch and reveal that the sampled representatives maximize a well-defined notion of data coverage in a transformed space. In addition, we present a highly scalable randomized implementation of the proposed algorithm shown to bring about substantial speedups. MOSAIC's superiority in achieving the desired characteristics of a representative subset all at once while exhibiting remarkable robustness to various outlier types is demonstrated via extensive experiments conducted on both real and synthetic data with comparisons to state-of-the-art algorithms.

Artificial intelligence (AI)-enhanced automated fault diagnosis (AFD) has become increasingly popular for chiller fault diagnosis with promising classification performance. In practice, a sufficient number of fault samples are required by the AI methods in the training phase. However, faulty training samples are generally much more difficult to be collected than normal training samples. Data augmentation is introduced in these scenarios to enhance the training data set with synthetic data. In this study, a variational autoencoder-based conditional Wasserstein GAN with gradient penalty (CWGAN-GP-VAE) is proposed to diagnose various faults for chillers. A detailed comparative study has been conducted with real-world fault data samples to verify the effectiveness and robustness of the proposed methodology.

Information extraction from synthetic aperture radar (SAR) images is heavily impaired by speckle noise, and hence, despeckling is a crucial preliminary step in scene analysis algorithms. The recent success of deep learning envisions a new generation of despeckling techniques that could outperform classical model-based methods. However, current deep learning approaches to despeckling require supervision for training, whereas clean SAR images are impossible to obtain. In the literature, this issue is tackled by resorting to either synthetically speckled optical images, which exhibit different properties with respect to true SAR images, or multitemporal SAR images, which are difficult to acquire or fuse accurately. In this article, inspired by recent works on blind-spot denoising networks, we propose a self-supervised Bayesian despeckling method. The proposed method is trained by employing only noisy SAR images and can, therefore, learn features of real SAR images rather than synthetic data. Experiments show that the performance of the proposed approach is very close to the supervised training approach on synthetic data and superior on real data in both quantitative and visual assessments.

Gravity and magnetic inversion are important methods for comprehensive quantitative interpretation of data obtained in, e.g., mineral, oil and gas, and geothermal exploration. At present, the 3-D joint inversion technology of gravity and magnetic data is facing challenges from large-scale data exploration applications. In this letter, a new algorithm for 3-D joint inversion of gravity and magnetic data with high accuracy and low computational cost is presented. We use the geometric trellis method to perform fast forward calculations and then introduce the sparse constraint and adaptive sensitivity matrix into the model constraint terms. The inexact structural resemblance method is then used to add the cross-gradient constraint penalty term to the objective function. Finally, an algorithm (DS-TGN) combining data-space (DS) and truncated Gauss-Newton (TGN) methods is used to solve the joint inversion objective function. Numerical experiments with synthetic data show that the proposed algorithm can significantly reduce the computational cost and obtain high accuracy density and magnetization models with structural resemblance and sharp boundaries. We also apply the DS-TGN algorithm to data obtained in the area of Greater Khingan in northwestern Heilongjiang, China. The underground density and magnetization distribution results provide a high-resolution geological model for the detection of skarn-type deposits.

Fault detection is a crucial task in seismic structure interpretation. Convolutional neural network (CNN)-based methods, in general, require large amount of labeled data for network training. One way to build the labeled data is to create synthetic seismic images with corresponding fault labels. However, it is hard to ensure that the synthetic data have the same fault feature distributions as the field data, which may lead to inaccurate and unreliable prediction results. Another way is to manually label the faults, which is time-consuming and subjective. In this letter, we propose that using knowledge distillation (KD) to improve the performance of fault detection by integrating the features from large number of synthetic samples and a small number of field samples. We distill knowledge from an ensemble of two teacher CNNs to train a student CNN (applied to final target) for seismic fault detection. In our work, one segmentation teacher CNN is trained on synthetic samples with known ground truth fault labels and another classification teacher CNN is trained on field samples with manually picked labels. Then, a classification student network is trained on samples generated by voting the results from two teacher models. The student CNN learns not only the general fault characteristics in the synthetic data but also the specific fault features of the target field data. Test on the field data shows that the student CNN highlights seismic fault more accurately with higher resolution than the teacher CNNs.

Images captured in foggy weather conditions suffers from inevitable problems such as low contrast, blurriness and less visibility. In many computer and vision applications like surveillance, object detection, object tracking and navigation, these low-quality images cannot be used and it requires improvement in the image quality. Different algorithms have been proposed in this direction and to upgrade the quality of a foggy image. Most of the existing methods gives good quality image but with high time complexity. In this paper, a novel and effective method is proposed to remove fog from a single image. The proposed method is based on principal component analysis and modified dark channel prior. In proposed algorithm, foggy image is pre-processed using principal component analysis. This pre-processed image is further enhanced using fast global smoothening filter. Time complexity of the proposed method is much less as compared to the various existing methods and at the same time, quality is also maintained. Also, proposed algorithm does not require a large data set and specific hardware. To see the effectiveness of the proposed technique, both qualitative and quantitative analysis has been done on synthetic data set as well as on natural dataset.

Spatio-temporal information processing is fundamental in both brain functions and AI applications. Current strategies for spatio-temporal pattern recognition usually involve explicit feature extraction followed by feature aggregation, which requires a large amount of labeled data. In the present study, motivated by the subcortical visual pathway and early stages of the auditory pathway for motion and sound processing, we propose a novel brain-inspired computational model for generic spatio-temporal pattern recognition. The model consists of two modules, a reservoir module and a decision-making module. The former projects complex spatio-temporal patterns into spatially separated neural representations via its recurrent dynamics, the latter reads out neural representations via integrating information over time, and the two modules are linked together using known examples. Using synthetic data, we demonstrate that the model can extract the frequency and order information of temporal inputs. We apply the model to reproduce the looming pattern discrimination behavior as observed in experiments successfully. Furthermore, we apply the model to the gait recognition task, and demonstrate that our model accomplishes the recognition in an event-based manner and outperforms deep learning counterparts when training data is limited.

Measured decay heat data of light water reactor (LWR) spent nuclear fuel (SNF) assemblies are adopted to train machine learning (ML) models. The measured data is available for fuel assemblies irradiated in commercial reactors operated in the United States and Sweden. The data comes from calorimetric measurements of discharged pressurized water reactor (PWR) and boiling water reactor (BWR) fuel assemblies. 91 and 171 measurements of PWR and BWR assembly decay heat data are used, respectively. Due to the small size of the measurement dataset, we propose: (i) to use the method of multiple runs (ii) to generate and use synthetic data, as large dataset which has similar statistical characteristics as the original dataset. Three ML models are developed based on Gaussian process (GP), support vector machines (SVM) and neural networks (NN), with four inputs including the fuel assembly averaged enrichment, assembly averaged burnup, initial heavy metal mass, and cooling time after discharge. The outcomes of this work are (i) development of ML models which predict LWR fuel assembly decay heat from the four inputs (ii) generation and application of synthetic data which improves the performance of the ML models (iii) uncertainty analysis of the ML models and their predictions.

Designing a generative model to synthesize realistic tabular data is of great significance in data science. Existing tabular data generative models have difficulty in handling complicated and diverse marginal distribution types due to the gradient vanishing problem, and these models pay little attention to the correlation between attributes. We propose a method that improves the generative adversarial network (GAN) with inverse cumulative distribution function for tabular data synthesis. This method first transforms continuous columns into uniform distribution data by using the cumulative distribution function, which can alleviate the gradient vanishing problem in model training. Then the method trains GAN with the transformed data, where the discriminator with label reconstruction function is presented to model the correlation among attributes accurately by introducing an auxiliary supervised task to help the correlations extraction. After that, we train a neural network for each continuous column to perform the inverse transformation of generated data into the target distribution, thereby the synthetic data is obtained. Experiments on simulated and real-world datasets show that our method compares favorably against the state-of-the-art methods in modeling tabular data.

Prescriptive analytics is a mechanism that provides the best set of actions to be taken to prevent undesirable outcomes for a given instance. However, this mechanism is prone to privacy breaches if an adversary with subsidiary data is allowed multiple query access to it. So, we propose a differential privacy mechanism in prescriptive analytics to preserve data privacy. Differential privacy can be achieved with the help of sensitivity of the given actions. Roughly speaking, sensitivity is the maximum change in the given set of actions with respect to the change in the given instances. However, a general analytical form for the sensitivity of the prescriptive analytics mechanism is difficult to derive. So, we formulate a nested constrained optimization to solve the problem. We use synthetic data in the experiments to validate the behavior of the differential privacy mechanism with respect to different privacy parameter settings. The experiments with two real-world datasets—Student Academic Performance and Reddit dataset, demonstrate the usefulness of our proposed method in education and social policy design. We also propose a new evaluation measure called the prescription success rate to further investigate the significance of our proposed method.

One of the many applications of X-ray computed tomography (CT) in industry is the detection of pores, cavities and other flaws in cast metal parts. Because of its improvement on part safety and saving of expenses, CT inspection is moving from a random sample inspection towards a full in-line inspection. With the increasing amount of produced data, however, comes the need for an automated processing. Due to tight time constraints the resulting CT scans are very artifact afflicted, which impedes automated inspection. In recent years, deep learning methods—convolutional neural networks in particular—have been used with great success to tackle even complex segmentation tasks in cluttered scenes. As we show, these methods are also applicable to the domain of industrial CT data: they are able to cope with noise, beam hardening, scatter and other artifacts which we encounter here. However, these methods need a vast amount of precisely labeled training data to work properly. Gathering the necessary data is not only cumbersome due to the need of annotating three-dimensional data but also expensive as it requires the knowledge of domain experts. Therefore, we present a new approach: We train our models on realistically simulated CT data only. Here, a precise per-voxel ground truth can simply be computed. In order to show that the simulated data is sufficient to train a segmentation network, we turn to its prediction performance on real CT data. We compare the prediction performance of traditional algorithms as well as the trained segmentation network on simulated and real validation data and demonstrate that they behave similarly. The ground truth for the real validation data is hand-labeled using high-quality CT scans, while the actual validation set consists of CT scans of lower quality of the exact same parts. For a comprehensive evaluation, we evaluate the probability of detection as well as the intersection over union. The first tells us how likely a flaw of given size can be found with a given confidence, which is of special interest to domain experts. The latter gives us a per-voxel information of how precise the overall segmentation is. Moreover, our synthetic data enables us to examine the influence of different artifact types on the detection rate. Besides these quantitative analyses we show some qualitative results of real-world applications. To the best of our knowledge, we describe the first approach for defect detection in three-dimensional CT data, which is solely trained with simulated data.

Satellite-derived geophysical variables provide valuable information about the earth’s functioning, but there are errors that limit their direct uses. Linearly combining two or more data sources is a simple and efficient method to reduce the uncertainty between the truth and observations. However, calculating the optimal weight for such a linear combination generally needs a reference “truth” that is rarely available in practical applications. To address this limitation, a triple collocation (TC) technique is often used to estimate data error by using a data triplet without the truth. The TC-based error variances are then applied to calculate the optimal weight calculation with an assumption of independence between errors. However, ignoring the error cross-correlations (ECCs) can lead to inaccurate optimal weight for a linear combination and hence may degrade the performance of calculations using merged products. Recently, an extended double instrumental variable (EIVD) method has been proposed to estimate ECC using only three products like TC. In this study, we examined the performance of an EIVD-based linear combination approach through two applications using synthetic data and actual satellite-derived soil moisture (SM) products. For direct comparison, the TC-based linear combination was also implemented using the same data sets. The verification showed that the EIVD-based products are generally better than the TC-based products and result in statistically significant differences in correlation coefficients. While the results here are based on SM, the proposed method is potentially applicable for reconstruction of other data sets that seldom have suitable references in large-scale regional settings.

International audience; Purpose: We propose to learn a 3D keypoint descriptor which we use to match keypoints extracted from full-body CT scans. Our methods are inspired by 2D keypoint descriptor learning, which was shown to outperform hand-crafted descriptors. Adapting these to 3D images is challenging because of the lack of labelled training data and high memory requirements.Method: We generate semi-synthetic training data. For that, we first estimate the distribution of local affine inter-subject transformations using labelled anatomical landmarks on a small subset of the database. We then sample a large number of transformations and warp unlabelled CT scans, for which we can subsequently establish reliable keypoint correspondences using guided matching. These correspondences serve as training data for our descriptor, which we represent by a CNN and train using the triplet loss with online triplet mining.Results: We carry out experiments on a synthetic data reliability benchmark and a registration task involving 20 CT volumes with anatomical landmarks used for evaluation purposes. Our learned descriptor outperforms the 3D-SURF descriptor on both benchmarks while having a similar runtime.Conclusion: We propose a new method to generate semi-synthetic data and a new learned 3D keypoint descriptor. Experiments show improvement compared to a hand-crafted descriptor. This is promising as literature has shown that jointly learning a detector and a descriptor gives further performance boost.

Location privacy-preserving mechanisms (LPPMs) have been extensively studied for protecting users' location privacy by releasing a perturbed location to third parties such as location-based service providers. However, when a user's perturbed locations are released continuously, existing LPPMs may not protect the sensitive information about the user's spatiotemporal activities, such as "visited hospital in the last week" or "regularly commuting between Address 1 and Address 2" (it is easy to infer that Addresses 1 and 2 may be home and office), which we call it \textit{spatiotemporal event}. In this paper, we first formally define {spatiotemporal event} as Boolean expressions between location and time predicates, and then we define $ \epsilon $-\textit{spatiotemporal event privacy} by extending the notion of differential privacy. Second, to understand how much spatiotemporal event privacy that existing LPPMs can provide, we design computationally efficient algorithms to quantify the privacy leakage of state-of-the-art LPPMs when an adversary has prior knowledge of the user's initial probability over possible locations. It turns out that the existing LPPMs cannot adequately protect spatiotemporal event privacy. Third, we propose a framework, PriSTE, to transform an existing LPPM into one protecting spatiotemporal event privacy against adversaries with \textit{any} prior knowledge. Our experiments on real-life and synthetic data verified that the proposed method is effective and efficient., Comment: accepted in TKDE. arXiv admin note: substantial text overlap with arXiv:1810.09152

The rapid aging of the population worldwide requires increased attention from healthcare providers and the entire society. For the elderly to live independently, many health issues related to old age, such as frailty and risk of falling, need increased attention and monitoring. When monitoring daily routines for older adults, it is desirable to detect the early signs of health changes before serious health events, such as hospitalizations, happen so that timely and adequate preventive care may be provided. By deploying multi-sensor systems in homes of the elderly, we can track trajectories of daily behaviors in a feature space defined using the sensor data. In this article, we investigate a methodology for tracking the evolution of the behavior trajectories over long periods (years) using high-dimensional streaming clustering and provide very early indicators of changes in health. If we assume that habitual behaviors correspond to clusters in feature space and diseases produce a change in behavior, albeit not highly specific, tracking trajectory deviations can provide hints of early illness. Retrospectively, we visualize the streaming clustering results and track how the behavior clusters evolve in feature space with the help of two dimension-reduction algorithms: Principal Component Analysis and t-distributed Stochastic Neighbor Embedding. Moreover, our tracking algorithm in the original high-dimensional feature space generates early health warning alerts if a negative trend is detected in the behavior trajectory. We validated our algorithm on synthetic data and tested it on a pilot dataset of four TigerPlace residents monitored with a collection of motion, bed, and depth sensors over 10 years. We used the TigerPlace electronic health records to understand the residents’ behavior patterns and to evaluate the health warnings generated by our algorithm. The results obtained on the TigerPlace dataset show that most of the warnings produced by our algorithm can be linked to health events documented in the electronic health records, providing strong support for a prospective deployment of the approach.

Practical application of deep learning based non-intrusive load monitoring (NILM) system requires the deep neural network model to generalize on new unseen data. Existing NILM solutions are not suitable for real-world application due to their poor disaggregation accuracy on new unseen data. In order to address this problem, this paper presents a NILM algorithm that uses data augmentation to generate synthetic data for training deep convolutional neural network models for each target appliance. Proposed data augmentation technique works by combining on and off-durations of a target appliance from various datasets, and forms a unified and comprehensive synthetic aggregate and sub-meter profiles. Apart from proposed algorithm, this paper also proposes an evaluation approach that relies on total predicted energy and ground-truth energy of an appliance to provide detailed insights about total overlapping energy, missing energy and extra energy predicted by the algorithm. Comparison results on our proposed evaluation approach showed that proposed disaggregation algorithm was able to predict energy that was 60% overlapping with ground truth energy and 36% energy was extra. Overall results showed that overlapping energy was 2.5 times more, and extra-predicted energy was 60% less than state-of-the-art algorithms in unseen test cases.

In photography, the presence of a bright light source often reduces the quality and readability of the resulting image. Light rays reflect and bounce off camera elements, sensor or diaphragm causing unwanted artifacts. These artifacts are generally known as "lens flare" and may have different influences on the photo: reduce contrast of the image (veiling glare), add circular or circular-like effects (ghosting flare), appear as bright rays spreading from light source (starburst pattern), or cause aberrations. All these effects are generally undesirable, as they reduce legibility and aesthetics of the image. In this paper we address the problem of removing or reducing the effect of veiling glare on the image. There are no available large-scale datasets for this problem and no established metrics, so we start by (i) proposing a simple and fast algorithm of generating synthetic veiling glare images necessary for training and (ii) studying metrics used in related image enhancement tasks (dehazing and underwater image enhancement). We select three such no-reference metrics (UCIQE, UIQM and CCF) and show that their improvement indicates better veil removal. Finally, we experiment on neural network architectures and propose a two-branched architecture and a training procedure utilizing structural similarity measure.

Infrared small-target detection under heterogeneous background, as a challenging task, plays an important role in many applications. In practice, there are not only bright targets but also dim targets, e.g., rescue aircraft and vehicles in the forest fire scene. Considering that most existing infrared small-target detection methods are merely aimed at bright targets, a novel method using multiple morphological profiles (MMP) is proposed, which can detect various types of targets whose brightness varies greatly. In the designed morphological feature extraction, different attributes, i.e., area attribute and height attribute, are applied to extract spatial size and contrast information of small-target in the max-tree and min-tree, respectively. Furthermore, discontinuous pruning values are further utilized for different attributes, and a designed fusion strategy of different pruning values results in more robust detection performance. Experimental results validated on two synthetic data and six real data sets demonstrate that the proposed MMP can not only detect a variety of brightness of targets and different types of targets and kinds of spatial sizes of targets but also further improve the contrast between targets and background, and the background clutter is significantly suppressed.

Generating computer graphics (CG) rendered synthetic images has been widely used to create simulation environments for robotics/autonomous driving and generate labeled data. Yet, the problem of training models purely with synthetic data remains challenging due to the considerable domain gaps caused by current limitations on rendering. In this paper, we propose a simple yet effective domain adaptation framework towards closing such gap at image level. Unlike many GAN-based approaches, our method aims to match the covariance of the universal feature embeddings across domains, making the adaptation a fast, convenient step and avoiding the need for potentially difficult GAN training. To align domains more precisely, we further propose a conditional covariance matching framework which iteratively estimates semantic segmentation regions and conditionally matches the class-wise feature covariance given the segmentation regions. We demonstrate that both tasks can mutually refine and considerably improve each other, leading to state-of-the-art domain adaptation results. Extensive experiments under multiple synthetic-to-real settings show that our approach exceeds the performance of latest domain adaptation approaches. In addition, we offer a quantitative analysis where our framework shows considerable reduction in Frechet Inception distance between source and target domains, demonstrating the effectiveness of this work in bridging the synthetic-to-real domain gap.

The latest industrial revolution, Industry 4.0, is progressing exponentially and targets to integrate artificial intelligence and machine learning algorithms with existing technology to digitalise chemical processes across the industry, especially in the area of online monitoring, predictive analysis and modelling. Machine learning algorithms are being constantly implemented in both academic laboratories and industry to uncover the underlying correlations that exist in the high-dimensional and complex experimental and synthetic data that describes a chemical process. Indeed soon, proficiency in artificial intelligence methodology will become a required skill of a chemical engineer. It is therefore becoming essential to train chemical engineers with these methods to help them to adapt to this new era of digitised industries. Keeping these issues in mind, we introduced deep neural networks to the final-year chemical engineering students through a computer laboratory exercise. The exercise was delivered in fast-track mode: the students were asked to develop deep neural networks to model and predict the equilibrium adsorption of uptake of three different acids by activated carbon at four different temperatures. In this manuscript, we discuss in detail this laboratory exercise from delivery and design to the results obtained and the students’ feedback. In the classroom, the students compared the adsorption equilibrium data obtained using the established theoretical adsorption isotherms and empirical correlations with the neural networks developed in the classroom. The experience obtained from the classroom confirmed that this exercise gave the students the essential knowledge on the AI and awareness on the jargons in the world of machine language and obtained the required level of coding skills to develop a simple neural net with one layer or a sophisticated deep networks to model an important unit operation in chemical engineering and to accurately predict the experimental outcomes.

Cyber-physical-social systems (CPSSs) handle large-scale multi-source data in different application areas, and the collected data usually contain personal private information and uncompleted information, which are typically distributed in different locations. Tensor completion has been widely used for recovering the missing entries in scale multidimensional data, and has proven to be an effective method. Privacy-preserving tensor completion in CPSSs, however, faces challenging issues, such as scalability, scatter, and security. In this paper, we propose a privacy-preserving tensor completion method that uses the optimized federated soft-impute algorithm with a differentially private guarantee. Moreover, we theoretically analyzed the privacy guarantee and utility guarantee. We evaluated the proposed algorithms on both synthetic data and real-world data. The results show that our algorithm performed better and provided strong privacy protection under a federated learning framework. Our method significantly saved space and time for privacy-preserving tensor completion in a CPSS.

Scalable training data generation is a critical problem in deep learning. We propose PennSyn2Real - a photo-realistic synthetic dataset consisting of more than 100 000 4K images of more than 20 types of micro aerial vehicles (MAVs). The dataset can be used to generate arbitrary numbers of training images for high-level computer vision tasks such as MAV detection and classification. Our data generation framework bootstraps chroma-keying, a mature cinematography technique, with a motion tracking system providing artifact-free and curated annotated images. Our system, therefore, allows object orientations and lighting to be controlled. This framework is easy to set up and can be applied to a broad range of objects, reducing the gap between synthetic and real-world data. We show that synthetic data generated using this framework can be directly used to train CNN models for common object recognition tasks such as detection and segmentation. We demonstrate competitive performance in comparison with training using only real images. Furthermore, bootstrapping the generated synthetic data in few-shot learning can significantly improve the overall performance, reducing the number of required training data samples to achieve the desired accuracy.

The human eye contains valuable information about an individual’s identity and health. Therefore, segmenting the eye into distinct regions is an essential step towards gathering this useful information precisely. The main challenges in segmenting the human eye include low light conditions, reflections on the eye, variations in the eyelid, and head positions that make an eye image hard to segment. For this reason, there is a need for deep neural networks, which are preferred due to their success in segmentation problems. However, deep neural networks need a large amount of manually annotated data to be trained. Manual annotation is a labor-intensive task, and to tackle this problem, we used data augmentation methods to improve synthetic data. In this paper, we detail the exploration of the scenario, which, with limited data, whether performance can be enhanced using similar context data with image augmentation methods. Our training and test set consists of 3D synthetic eye images generated from the UnityEyes application and manually annotated real-life eye images, respectively. We examined the effect of using synthetic eye images with the Deeplabv3+ network in different conditions using image augmentation methods on the synthetic data. According to our experiments, the network trained with processed synthetic images beside real-life images produced better mIoU results than the network, which only trained with real-life images in the Base dataset. We also observed mIoU increase in the test set we created from MICHE II competition images.

Reconstructing hand-object interactions is a challenging task due to strong occlusions and complex motions. This article proposes a real-time system that uses a single depth stream to simultaneously reconstruct hand poses, object shape, and rigid/non-rigid motions. To achieve this, we first train a joint learning network to segment the hand and object in a depth image, and to predict the 3D keypoints of the hand. With most layers shared by the two tasks, computation cost is saved for the real-time performance. A hybrid dataset is constructed here to train the network with real data (to learn real-world distributions) and synthetic data (to cover variations of objects, motions, and viewpoints). Next, the depth of the two targets and the keypoints are used in a uniform optimization to reconstruct the interacting motions. Benefitting from a novel tangential contact constraint, the system not only solves the remaining ambiguities but also keeps the real-time performance. Experiments show that our system handles different hand and object shapes, various interactive motions, and moving cameras.

In impedance cardiography (ICG), the detection of dZ/dt signal (ICG) characteristic points, especially the X point, is a crucial step for the calculation of hemodynamic parameters such as stroke volume (SV) and cardiac output (CO). Unfortunately, for beat-to-beat calculations, the accuracy of the detection is affected by the variability of the ICG complex subtypes. Thus, in this work, automated classification of ICG complexes is proposed to support the detection of ICG characteristic points and the extraction of hemodynamic parameters according to several existing subtypes. A novel pattern recognition artificial neural network (PRANN) approach was implemented, and a divide-and-conquer strategy was used to identify the five different waveforms of the ICG complex waveform with output nodes no greater than 3. The PRANN was trained, tested and validated using a dataset from four volunteers from a measurement of eight electrodes. Once the training was satisfactory, the deployed network was validated on two other datasets that were completely different from the training dataset. As an additional performance validation of the PRANN, each dataset included four volunteers for a total of eight volunteers. The results show an average accuracy of 96% in classifying ICG complex subtypes with only a decrease in the accuracy to 83 and 80% on the validation datasets. This work indicates that the PRANN is a promising method for automated classification of ICG subtypes, facilitating the investigation of the extraction of hemodynamic parameters from beat-to-beat dZ/dt complexes.

This study aims to predict the pavement layer moduli using responses from surface deflections data using a lightweight deflectometer (LWD). For thin asphalt pavement, LWD may be a useful tool for evaluation purposes. A finite element method, Plaxis 2D software, was used to analyse the LWD stress influences on selected pavement crust thickness. Conversion of in situ deflections values into moduli values is known as the backcalculation technique and; it is a complex and mathematical process that relies solely on one of the factors of measured surface deflections under a given load. It is challenging to measure backcalculate moduli of thin pavements. Thus, this study aims to develop an empirical equation using regression analysis for a quick approximate estimation of in situ layer moduli using user-friendly interference Microsoft excel without any seed value and software interface requirement. In this regard, Kenlayer pavement analysis was used to develop synthetic data, which helps develop an empirical equation. Validation for the developed empirical model deals with the genetic algorithm base backcalculation approach, laboratory results and field test data. Validation of the results strengthens the developed empirical equation. Eventually, LWD with 20 kg drop weight proposed empirical model provides a quick approximate estimation, and efficient way to access the existing thin asphalt pavement condition may be the further no requirement of backcalculation process.

At present, most of the seismic data denoising methods based on deep learning attempt to establish a synthetic seismic data set as the network training set to train network parameters. However, the synthetic data set cannot completely reflect the structural characteristics of the field seismic data, resulting in some false seismic reflections in field denoised results. For this reason, this article proposes an attribute-based denoising algorithm for seismic data called attribute-based double constraint denoising network (Att-DCDN). This method applies encoder–decoder and attribute classifier to constitute the generative adversarial network (GAN) and attenuates seismic noise by controlling with/without target attributes (noise attribute and signal attribute). Compared with the noise-free field seismic data, attribute vectors of the field data are easier to obtain. Therefore, our training set includes not only the synthetic seismic data but also the field seismic data, so as to reduce accuracy requirement of the synthetic noise-free data. In addition, we propose a double-constraint training way to reduce the losses of effective reflections during the denoising process. Specifically, we consider both noise attenuation and signal retention, i.e., reconstruction loss and residual loss are introduced to constrain recovery of effective reflections, and attribute classification loss and adversarial loss are applied to constrain the attenuation of seismic noise. Both the experimental results of synthetic and field seismic records show that our algorithm can effectively suppress the seismic noise and recover the effective reflections almost completely, even the weak signal areas that are seriously polluted by the seismic noise.

This paper presents a data-driven measurement model for extended object tracking (EOT) with automotive radar. Specifically, the spatial distribution of automotive radar measurements is modeled as a hierarchical truncated Gaussian (HTG) with structural geometry parameters that can be learned from the training data. The HTG measurement model provides an adequate resemblance to the spatial distribution of real-world automotive radar measurements. Moreover, large-scale radar datasets can be leveraged to learn the geometry-related model parameters and offload the computationally demanding model parameter estimation from the state update step. The learned HTG measurement model is further incorporated into a random matrix based EOT approach with two (multi-sensor) measurement updates: one is based on a factorized Gaussian inverse-Wishart density representation and the other is based on a Rao-Blackwellized particle density representation. The effectiveness of the proposed approaches is verified on both synthetic data and real-world nuScenes dataset over 300 trajectories.

Thousands of high-speed trains (HSTs) run on over 30000-km high-speed railway every day in China. These running trains not only cause vibrations but also stimulate all kinds of seismic waves. The receiver seismic signal’s spectrum has one special equal-interval, multinarrowband feature. Some commonly used time-frequency (TF) analyzing methods, such as the short-time Fourier transform, continuous wavelet transform, and the Stockwell transform, can capture a signal’s TF features and reconstruct signal without loss. However, their TF resolution is limited by the uncertainty principle. In this letter, we apply the second-order synchrosqueezing continuous wavelet transform (SSCWT), which is one improved SSCWT, to the HST induced seismic signal and capture its TF spectrum features. The synthetic data example and real data example have proved the second-order SSCWT can provide a more concentrated TF spectrum than the commonly used SSCWT.

3D point cloud denoising is an increasingly demanding field as such type of data structure is getting more attention in perceiving the 3D environment for diverse applications. Despite their novelty, recently proposed solutions are still modest in terms of effectiveness and robustness, especially for scenes corrupted with a massive amount of noise. The encountered challenges are mainly due to the data acquisition process and the little-to-no knowledge of the statistical data distribution. In this paper, two promising unsupervised machine learning techniques are investigated, which are the Isolation Forest (If) and the Elliptic Envelope (EE). Each of these techniques detects noise using different philosophies. If uses a forest of iTrees; while EE uses a learned imaginary elliptic. The proposed solution, named Point-Denoise, tunes both techniques and fuses them at the decision-level. Although the solution simplicity, Point-Denoise reports superior results to state-of-the-art techniques. For evaluation purposes, both synthetic and real data are used. The chosen synthetic data is the ModelNet40 benchmark, which is augmented with a Gaussian and emulated 3D scanner noise with three different standard deviations: 0.5%, 1.0%, and 1.5% assessing the robustness of the proposed methodology. Meanwhile, the real data is collected from the Qatar University campus. Considering that a massive amount of noise already corrupts real data at acquisition time, no additional noise is augmented. Point-Denoise outperforms state-of-the-art solutions (i.e., traditional filtering, supervised, and unsupervised learning techniques) by attaining a 0.24 distance error and achieving a 48.93% enhancement.

Let a class of proper curves is specified by positive examples only. We aim to propose a learning novelty detection algorithm that decides whether a new curve is outside this class or not. In opposite to the majority of the literature, two sources of a curve variability are present, namely, the one inherent to curves from the proper class and observations errors’. Therefore, firstly a decision function is trained on historical data, and then, descriptors of each curve to be classified are learned from noisy observations.When the intrinsic variability is Gaussian, a decision threshold can be established from T 2 Hotelling distribution and tuned to more general cases. Expansion coefficients in a selected orthogonal series are taken as descriptors and an algorithm for their learning is proposed that follows nonparametric curve fitting approaches. Its fast version is derived for descriptors that are based on the cosine series. Additionally, the asymptotic normality of learned descriptors and the bound for the probability of their large deviations are proved. The influence of this bound on the decision threshold is also discussed.The proposed approach covers curves described as functional data projected onto a finite-dimensional subspace of a Hilbert space as well a shape sensitive description of curves, known as square-root velocity (SRV). It was tested both on synthetic data and on real-life observations of the COVID-19 growth curves.

Real social network feature many small groups with information exchange among groups. Groups often overlap with each other, and individuals can be in multiple groups. By simulating the information transmission process in different networks with structural changes, we can obtain a deeper understanding of the attribute characteristics of complex networks and the law of information transmission. Therefore, to study the process of information exchange and dissemination in complex and overlapping group networks, this paper first constructs the friend circles network (FCnet) and then proposes a dynamic information dissemination circle-susceptible–infected–removed (C-SIR) model. In this model, we discuss the extent of interest and impact among neighbours in the same circle and cross circles in FCnet, as well as the scale of imminent connections between circles. The validity and authenticity of the model are verified based on synthetic data and real data, and the conclusion that the C-SIR model can accelerate information propagation, maintain popularity, and expand the infection radius is obtained.

Object detection typically assumes that training and test samples are drawn from an identical distribution, which, however, does not always hold in practice. Such a distribution mismatch may lead to a significant performance drop. In this work, we present Scale-aware Domain Adaptive Faster R-CNN, a model aiming at improving the cross-domain robustness of object detection. In particular, our model improves the traditional Faster R-CNN model by tackling the domain shift on two levels: (1) the image-level shift, such as image style, illumination, etc., and (2) the instance-level shift, such as object appearance, size, etc. The two domain adaptation modules are implemented by learning domain classifiers in an adversarial training manner. Moreover, we observe that the large variance in object scales often brings a crucial challenge to cross-domain object detection. Thus, we improve our model by explicitly incorporating the object scale into adversarial training. We evaluate our proposed model on multiple cross-domain scenarios, including object detection in adverse weather, learning from synthetic data, and cross-camera adaptation, where the proposed model outperforms baselines and competing methods by a significant margin. The promising results demonstrate the effectiveness of our proposed model for cross-domain object detection. The implementation of our model is available at https://github.com/yuhuayc/sa-da-faster .

Learning discriminative representations with deep neural networks often relies on massive labeled data, which is expensive and difficult to obtain in many real scenarios. As an alternative, self-supervised learning that leverages input itself as supervision is strongly preferred for its soaring performance on visual representation learning. This paper introduces a contrastive self-supervised framework for learning generalizable representations on the synthetic data that can be obtained easily with complete controllability. Specifically, we propose to optimize a contrastive learning task and a physical property prediction task simultaneously. Given the synthetic scene, the first task aims to maximize agreement between a pair of synthetic images generated by our proposed view sampling module, while the second task aims to predict three physical property maps, i.e., depth, instance contour maps, and surface normal maps. In addition, a feature-level domain adaptation technique with adversarial training is applied to reduce the domain difference between the realistic and the synthetic data. Experiments demonstrate that our proposed method achieves state-of-the-art performance on several visual recognition datasets.

Non-invasive extraction of fetal electrocardiogram (FECG) by processing the abdominal signals is emerging as a promising approach in the areas of obstetrics and gynecology. This paper presents a two-stage improved non-linear adaptive filter for FECG extraction. The reference input to the adaptive noise canceler (ANC) is first processed using an adaptive neuro-fuzzy inference system (ANFIS) to estimate the non-linear maternal component in abdominal signals. A parallel sub-filter (PSF) ANC is proposed to assess the fetal ECG from the abdominal signal. The PSF-ANC decomposes a single adaptive filter into multiple sub-filters to improve the convergence performance. The filter coefficients of PSF-ANC adaptively obtained using normalised least mean square algorithm by minimizing the mean square error. Different error and common error algorithms are proposed based on the computation of the error signal. A synthetic data from the FECG synthetic database is used to evaluate the convergence performance. Two real-time data from the Daisy database and the Non-invasive FECG database from Physionet are used to evaluate the proposed ANFIS-PSF’s performance qualitative and quantitatively. The results justify the performance improvement of proposed ANFIS-PSF ANC compared to the state of art techniques. The proposed scheme achieves a sensitivity of 97.92%, 94.52% accuracy, a positive predictive value of 94.66%, and an F1 score of 96.12%.

In this work, we deal with the problem of atmospheric compensation (AC) of hyperspectral data collected in the visible and near-infrared (VNIR) spectral range. We propose the “learning-based” approach which uses artificial intelligence algorithms to directly estimate the spectral reflectance from the observed at-sensor radiance image. It uses a parametric regressor whose parameters are learned by means of a strategy based on synthetic data. Such data are generated taking into account 1) the radiative transfer in the atmosphere; 2) the variability of the surface spectral reflectance; and 3) the effects of signal-dependent random noise and spectral miscalibration errors. According to this general framework, we propose a specific multilinear regressor that starting from the knowledge of the atmospheric visibility compensates the water absorption and provides the spectral reflectance of each pixel of the analyzed image. Furthermore, a specific image-based procedure is presented for visibility estimation. The experiment over simulated data is presented and discussed. The test on simulated data aims at showing the effectiveness of the proposed strategy in a completely controlled environment. Experiments are also carried out on three real hyperspectral images acquired by two hyperspectral sensors. The obtained results confirm the effectiveness of the proposed approach by comparing the retrieved reflectance spectra with in-situ measurements or with those obtained by using a well-known commercial AC software.

Recently, due to its widespread applications in public safety, anomaly detection in crowd scenes has become a hot topic. Some deep-learning-based methods attain significant achievements in this field. Nevertheless, most of them suffer from over-fitting to some extent because of scarce data, which are usually abrupt and low-frequency in the real world. To remedy the above problem, this paper firstly develops a synthetic anomaly event generating system, which could simulate typical specific abnormal events. By utilizing this system, a large synthetic, diverse anomaly event dataset is built, which contains 2,149 video sequences. After getting the dataset, a 3D CNN is designed to detect the abnormal types at the video level. However, we find that there are obvious domain differences (also named as “domain gap/shifts”) between synthetic videos and real-world data, which results in performance degradation when applying the model to the real world. Thus, this paper further proposes a cyclic 3D GAN for domain adaption to reduce the domain gap, which translates the synthetic data to the photorealistic video sequences. Then the detection model is trained on the translated data and it can perform well in the real data. Experimental results illustrate that the proposed method outperforms these baselines for the domain adaptation anomaly detection.

The self-expressiveness property of the data, i.e., each sample is linearly represented as a combination of other samples, has recently aroused much attention in the community of data mining and machine learning and shown great promising for subspace clustering. However, real-world data are usually contaminated by noise and outliers. The true similarity between samples directly learned from the original data may deviate from the intrinsic structure of the data, and subsequent clustering results will be severely affected. Hence naively taking a corrupted dictionary, i.e., data itself, will not always obtain the desired clustering performance. To address the above issues, this paper proposes a robust principal component analysis induced self-representation clustering method via adaptively learning the similarity graph from the clean data in the self-representation framework, where the original data are recovered, and the representation coefficients are obtained simultaneously in a unified framework. Specifically, we jointly integrate a nonnegative constraint and a distance regularization into the proposed framework, which guarantees the learned affinity matrix can simultaneously capture the global and local structure of data. Experimental results on both synthetic data and several famous multimedia datasets demonstrate that the proposed method performs much better against state-of-the-arts.

In this work, we propose a novel approach for the synthetization of data based on copulas, which are interpretable and robust models, extensively used in the actuarial domain. More precisely, our method COPULA-SHIRLEY is based on the differentially-private training of vine copulas, which are a family of copulas allowing to model and generate data of arbitrary dimensions. The framework of COPULA-SHIRLEY is simple yet flexible, as it can be applied to many types of data while preserving the utility as demonstrated by experiments conducted on real datasets. We also evaluate the protection level of our data synthesis method through a membership inference attack recently proposed in the literature.

In this paper, we propose a novel continuous latent semantic analysis fitting method, to efficiently and effectively estimate the parameters of model instances in data, based on latent semantic analysis and continuous preference analysis. Specifically, we construct a new latent semantic space (LSS): where inliers of different model instances are mapped into several independent directions, while gross outliers are distributed close to the origin of LSS. After that, we analyze the data distribution to effectively remove gross outliers in LSS, and propose an improved clustering algorithm to segment the remaining data points. On the one hand, the proposed fitting method is able to achieve excellent fitting results; due to the effective continuous preference analysis in LSS. On the other hand, the proposed method can efficiently obtain final fitting results due to the dimensionality reduction in LSS. Experimental results on both synthetic data and real images demonstrate that the proposed method achieves significant superiority over several state-of-the-art model fitting methods on both fitting accuracy and computational speed.

Purpose: Tracking of tools and surgical activity is becoming more and more important in the context of computer assisted surgery. In this work, we present a data generation framework, dataset and baseline methods to facilitate further research in the direction of markerless hand and instrument pose estimation in realistic surgical scenarios. Methods: We developed a rendering pipeline to create inexpensive and realistic synthetic data for model pretraining. Subsequently, we propose a pipeline to capture and label real data with hand and object pose ground truth in an experimental setup to gather high-quality real data. We furthermore present three state-of-the-art RGB-based pose estimation baselines. Results: We evaluate three baseline models on the proposed datasets. The best performing baseline achieves an average tool 3D vertex error of 16.7 mm on synthetic data as well as 13.8 mm on real data which is comparable to the state-of-the art in RGB-based hand/object pose estimation. Conclusion: To the best of our knowledge, we propose the first synthetic and real data generation pipelines to generate hand and object pose labels for open surgery. We present three baseline models for RGB based object and object/hand pose estimation based on RGB frames. Our realistic synthetic data generation pipeline may contribute to overcome the data bottleneck in the surgical domain and can easily be transferred to other medical applications.

Benchmarking unsupervised outlier detection is difficult. Outliers are rare, and existing benchmark data contains outliers with various and unknown characteristics. Fully synthetic data usually consists of outliers and regular instances with clear characteristics and thus allows for a more meaningful evaluation of detection methods in principle. Nonetheless, there have only been few attempts to include synthetic data in benchmarks for outlier detection. This might be due to the imprecise notion of outliers or to the difficulty to arrive at a good coverage of different domains with synthetic data. In this work, we propose a generic process for the generation of datasets for such benchmarking. The core idea is to reconstruct regular instances from existing real-world benchmark data while generating outliers so that they exhibit insightful characteristics. We propose and describe a generic process for the benchmarking of unsupervised outlier detection, as sketched so far. We then describe three instantiations of this generic process that generate outliers with specific characteristics, like local outliers. To validate our process, we perform a benchmark with state-of-the-art detection methods and carry out experiments to study the quality of data reconstructed in this way. Next to showcasing the workflow, this confirms the usefulness of our proposed process. In particular, our process yields regular instances close to the ones from real data. Summing up, we propose and validate a new and practical process for the benchmarking of unsupervised outlier detection.

In this article, we aim to develop a unified view of causal and non-causal feature selection methods. The unified view will fill in the gap in the research of the relation between the two types of methods. Based on the Bayesian network framework and information theory, we first show that causal and non-causal feature selection methods share the same objective. That is to find the Markov blanket of a class attribute, the theoretically optimal feature set for classification. We then examine the assumptions made by causal and non-causal feature selection methods when searching for the optimal feature set, and unify the assumptions by mapping them to the restrictions on the structure of the Bayesian network model of the studied problem. We further analyze in detail how the structural assumptions lead to the different levels of approximations employed by the methods in their search, which then result in the approximations in the feature sets found by the methods with respect to the optimal feature set. With the unified view, we can interpret the output of non-causal methods from a causal perspective and derive the error bounds of both types of methods. Finally, we present practical understanding of the relation between causal and non-causal methods using extensive experiments with synthetic data and various types of real-world data Refereed/Peer-reviewed

Sequentially detecting multiple changepoints in a data stream is a challenging task. Difficulties relate to both computational and statistical aspects, and in the latter, specifying control parameters is a particular problem. Choosing control parameters typically relies on unrealistic assumptions, such as the distributions generating the data, and their parameters, being known. This is implausible in the streaming paradigm, where several changepoints will exist. Further, current literature is mostly concerned with streams of continuous-valued observations, and focuses on detecting a single changepoint. There is a dearth of literature dedicated to detecting multiple changepoints in transition matrices, which arise from a sequence of discrete states. This paper makes the following contributions: a complete framework is developed for adaptively and sequentially estimating a Markov transition matrix in the streaming data setting. A change detection method is then developed, using a novel moment matching technique, which can effectively monitor for multiple changepoints in a transition matrix. This adaptive detection and estimation procedure for transition matrices, referred to as ADEPT-M, is compared to several change detectors on synthetic data streams, and is implemented on two real-world data streams – one consisting of over nine million HTTP web requests, and the other being a well-studied electricity market data set.

It is highly desirable to have a reliable modal parameter estimation technique with output-only measurements, especially for tall structures subjected to ambient wind loads. A modal identification technique called second-order blind identification (SOBI) is discussed in this paper to estimate the modal parameters of tall structures subjected to ambient loading. This method requires only acceleration time history data. The data are first preprocessed through cross-correlations to obtain the free response. SOBI uses these free responses to generate the mixing matrix (i.e., modal matrix) and sources, which are the modal responses of the structure. Even though SOBI has earlier been employed for modal parameter estimation, the majority of works attempt to solve small academic problems. Keeping this in view, we attempt to solve some practical engineering problems associated with tall structures. The proposed modal estimation technique is first evaluated using synthetic data of a 76-story tall benchmark building with varying levels of noise and later validated with the experimental data of the 610 m tall Guangzhou New TV Tower (GNTVT). Investigations carried out in this paper, using numerical and experimental ambient responses of tall structures, clearly indicate that the modal parameters estimated using SOBI are comparing very well with theoretical values.

Object pose estimation is a prerequisite for many robotic applications. Preparing dataset for network training is a challenging part of the pose estimation approaches, and in most of them, the network can detect just the trained objects. Synthetic data are used to train deep neural networks in robotic manipulation as a promising method for obtaining a huge amount of prelabeled training data, which are generated safely. We are to investigate the reality gap in the pose estimation of intra-category objects from a single RGB-D image using keypoints. The proposed approach in this paper provides a fast and simple procedure for training a deep neural network to identify the object and its keypoints based on synthetic dataset and autolabeling program. To our knowledge, this is the first deep network trained only on synthetic data that can find keypoints of intra-category objects for pose estimation purposes. The speed of training and the simplicity of this method make it very easy to add a new class of objects to the system which is the main advantage of this approach. Using this approach, we demonstrate a near-real-time system estimating object poses with sufficient accuracy for real-world semantic grasping and manipulating of intra-category objects in clutter by a real robot.

Synthetic data augmentation is of paramount importance for machine learning classification, particularly for biological data, which tend to be high dimensional and with a scarcity of training samples. The applications of robotic control and augmentation in disabled and able-bodied subjects still rely mainly on subject-specific analyses. Those can rarely be generalised to the whole population and appear to over complicate simple action recognition such as grasp and release (standard actions in robotic prosthetics and manipulators). We show for the first time that multiple GPT-2 models can machine-generate synthetic biological signals (EMG and EEG) and improve real data classification. Models trained solely on GPT-2 generated EEG data can classify a real EEG dataset at 74.71% accuracy and models trained on GPT-2 EMG data can classify real EMG data at 78.24% accuracy. Synthetic and calibration data are then introduced within each cross validation fold when benchmarking EEG and EMG models. Results show algorithms are improved when either or both additional data are used. A Random Forest achieves a mean 95.81% (1.46) classification accuracy of EEG data, which increases to 96.69% (1.12) when synthetic GPT-2 EEG signals are introduced during training. Similarly, the Random Forest classifying EMG data increases from 93.62% (0.8) to 93.9% (0.59) when training data is augmented by synthetic EMG signals. Additionally, as predicted, augmentation with synthetic biological signals also increases the classification accuracy of data from new subjects that were not observed during training. A Robotiq 2F-85 Gripper was finally used for real-time gesture-based control, with synthetic EMG data augmentation remarkably improving gesture recognition accuracy, from 68.29% to 89.5%.

An active area of research in computational science is the design of algorithms for solving the subgraph matching problem to find copies of a given template graph in a larger world graph. Prior works have largely addressed single-channel networks using a variety of approaches. We present a suite of filtering methods for subgraph isomorphisms for multiplex networks (with different types of edges between nodes and more than one edge within each channel type). We aim to understand the entire solution space rather than focusing on finding one isomorphism. Results are shown on several classes of datasets: (a) Sudoku puzzles mapped to the subgraph isomorphism problem, (b) Erdős-Rényi multigraphs, (c) real-world datasets from Twitter and transportation networks, (d) synthetic data created for the DARPA MAA program.

The study is concerned with a description of large numeric data with the aid of building a limited collection of representative information granules with the objective of capturing the structure of the original data. The proposed development scheme consists of two steps. First, a clustering algorithm characterized by high flexibility of coping with the diverse geometry of data structure and efficient computational overhead is invoked. At the second step, a clustering algorithm applied to the clusters already formed during the first phase, yielding a collection of numeric prototypes is involved and the numeric prototypes produced there are then generalized into their granular prototypes. The quality of granular prototypes is quantified while their build-up is supported by the mechanisms of granular computing such as the principle of justifiable granularity. In this paper, the clustering algorithms of DBSCAN and fuzzy ${C}$ -means were used in successive phases of the processed approach. The experimental studies concerning synthetic data and publicly available data are covered and the performance of the developed approach is assessed along with a comparative analysis.