Static Datasets Research Articles

Review: Outlier Detection using Clustering Based Techniques in Data Stream

Outlier detection is a critical data mining task used to detect and, where necessary, delete abnormal occurrences from dataset or tries to uncover useful anomalous and uneven patterns contained in large datasets. Various approaches are used to assign data points in different clusters based on different parameters. In this paper, we present a review of various existing approaches based on clustering of data points for detecting outliers from data set. The survey will cover the traditional outlier detection methods for static datasets, low dimensional datasets and recent developments that deal with outlier detection problems for dynamic data stream and high-dimensional datasets. In a comparative analysis, we highlight their benefits and drawbacks. Key Words: Outlier detection, Clustering based outlier detection, D-Stream, CORM, Hy-CARCE

Prevention of False Data Injection in Dataset using ML and DL

IoT is an interconnected and allotted network of embedded systems communicating through wired and wireless communication technology. It is defined as network of physical objects or things empowered with limited computation storage and conversation capabilities as well as embedded with electronics (sensors and actuators). Nowadays there are literally masses of hundreds of Internet of things (IoT) gadgets easily available to the customers. those consist of security cameras, smart home and smart speaker systems, smart toys and infant monitors, drones, domestic appliances, routers and internet gateways, and basically some other hardware products which can transmit information and be controlled over the net. FDIA is an attack this will result in a catastrophic outcomes. False information injection attack can be executed in each dynamic and static datasets, as a present device we easily discover and prevent in a static environment. in which with the present machine even the dynamic records is made static after which detects FDIA’s. FDIA in a static environment is the existing machine, in which we seek to have few solutions for the FDIA in a dynamic environment or for time collection information. So, to enhance protection in dynamic environment is our proposed challenge towards FDIA and we create a brand-new version known as “PdD” model – Predictive Dynamic version. Proposed framework performs crucial role in live streaming records in terms of heterogeneous environment (dynamic nature). So, we use GRU set of rules because it offers better RMSE value than different AI algorithms. This detects every time there may be a few FDI attack.

The orchestration of Machine Learning frameworks with data streams and GPU acceleration in Kafka‐ML: A deep‐learning performance comparative

AbstractMachine Learning (ML) applications need large volumes of data to train their models so that they can make high‐quality predictions. Given digital revolution enablers such as the Internet of Things (IoT) and the Industry 4.0, this information is generated in large quantities in terms of continuous data streams and not in terms of static datasets as it is the case with most AI (Artificial Intelligence) frameworks. Kafka‐ML is a novel open‐source framework that allows the complete management of ML/AI pipelines through data streams. In this article, we present new features for the Kafka‐ML framework, such as the support for the well‐known ML/AI framework PyTorch, as well as for GPU acceleration at different points along the pipeline. This pipeline will be described by taking a real Industry 4.0 use case in the Petrochemical Industry. Finally, a comprehensive evaluation with state‐of‐the‐art deep learning models will be carried out to demonstrate the feasibility of the platform.

EVALUATION OF SIMULTANEOUS IDENTITY, AGE AND GENDER RECOGNITION FOR CROWD FACE MONITORING

Nowadays, facial recognition combined with age estimation and gender prediction has been deeply involved with the factors associated with crowd monitoring. This is considered to be a major and complex job for humans. This paper proposes a unified facial recognition system based on already available deep learning and machine learning models (i.e., FaceNet, ResNet, Support Vector Machine, AgeNet and GenderNet) that automatically and simultaneously performs person identification, age estimation and gender prediction. Then the system is evaluated on a newly proposed multi-face, realistic and challenging test dataset. The current face recognition technology primarily focuses on static datasets of known identities and does not focus on novel identities. This approach is not suitable for continuous crowd monitoring. In our proposed system, whenever novel identities are found during inference, the system will save those novel identities with an appropriate label for each unique identity and the system will be updated periodically in order to correctly recognise those identities in the future inference iterations. However, extracting the facial features of the whole dataset whenever a new identity is detected is not an efficient solution. To address this issue, we propose an incremental feature extraction based training method which aims to reduce the computational load of feature extraction. When tested on the proposed test dataset, our proposed system correctly recognizes pre-trained identities, estimates age, and predicts gender with an average accuracy of 49%, 66.5% and 93.54% respectively. We conclude that the evaluated pre-trained models can be sensitive and not robust to uncontrolled environment (e.g., abrupt lighting conditions).

A Distributed Spatial Index With High Update Efficiency for Location-Based Real-Time Services

LBS-RT (location-based service in a real-time manner) has become popular because it can provide quick and timely services. Range query is often used in LBS-RT, which finds objects in a specified area, and spatial indices are often used to speed up range query. However, in LBS-RT, there are some difficulties. Spatial index was originally designed to index static dataset, but the dataset is dynamic in LBS-RT, which needs lots of insert and delete operations. To meet the gap, this paper proposes a new distributed spatial index called GQ-QBS. It's a two-layer master-slave mode that consists of a global index and multiple local indices. The global index (GQ-tree) is responsible for the dynamic load balancing and auto-scaling, while the local index (QBS-tree) is for quickly updating and querying. Experiments show the index has a significant advantage in LBS-RT.

The impact of inconsistent human annotations on AI driven clinical decision making

In supervised learning model development, domain experts are often used to provide the class labels (annotations). Annotation inconsistencies commonly occur when even highly experienced clinical experts annotate the same phenomenon (e.g., medical image, diagnostics, or prognostic status), due to inherent expert bias, judgments, and slips, among other factors. While their existence is relatively well-known, the implications of such inconsistencies are largely understudied in real-world settings, when supervised learning is applied on such ‘noisy’ labelled data. To shed light on these issues, we conducted extensive experiments and analyses on three real-world Intensive Care Unit (ICU) datasets. Specifically, individual models were built from a common dataset, annotated independently by 11 Glasgow Queen Elizabeth University Hospital ICU consultants, and model performance estimates were compared through internal validation (Fleiss’ κ = 0.383 i.e., fair agreement). Further, broad external validation (on both static and time series datasets) of these 11 classifiers was carried out on a HiRID external dataset, where the models’ classifications were found to have low pairwise agreements (average Cohen’s κ = 0.255 i.e., minimal agreement). Moreover, they tend to disagree more on making discharge decisions (Fleiss’ κ = 0.174) than predicting mortality (Fleiss’ κ = 0.267). Given these inconsistencies, further analyses were conducted to evaluate the current best practices in obtaining gold-standard models and determining consensus. The results suggest that: (a) there may not always be a “super expert” in acute clinical settings (using internal and external validation model performances as a proxy); and (b) standard consensus seeking (such as majority vote) consistently leads to suboptimal models. Further analysis, however, suggests that assessing annotation learnability and using only ‘learnable’ annotated datasets for determining consensus achieves optimal models in most cases.

INCORPORATING A HONEYFARM WITH MLFFNN IDS FOR IMPROVING INTRUSION DETECTION

Today’s networks must deal with dynamically changing threats each day. Use of static datasets to train and prepare multi-layer feed forward neural network intrusion detection systems (MLFFNN IDS) doesn’t address these new threats. The use of real traffic data to train neural network IDSs has been out of reach in organizations due to privacy and concerns. Now the use of a honeyfarm system can provide real-time data to a MLFFNN IDS so that it can adjust to new threats as they begin. This system also removes the privacy and concerns since information about the network is false and acts as a decoy to lure attackers away from the real organizational network. This paper introduces a honeyfarm architecture one can use with a MLFFNN IDS to improve intrusion detection capability.

Application of a satellite-retrieved sheltering parameterization (v1.0) for dust event simulation with WRF-Chem v4.1

Abstract. Roughness features (e.g., rocks, vegetation, furrows) that shelter or attenuate wind flow over the soil surface can considerably affect the magnitude and spatial distribution of sediment transport in active aeolian environments. Existing dust and sediment transport models often rely on vegetation attributes derived from static land use datasets or remotely sensed greenness indicators to incorporate sheltering effects on simulated particle mobilization. However, these overly simplistic approaches do not represent the three-dimensional nature or spatiotemporal changes of roughness element sheltering. They also ignore the sheltering contribution of non-vegetation roughness features and photosynthetically inactive (i.e., brown) vegetation common to dryland environments. Here, we explore the use of a novel albedo-based sheltering parameterization in a dust transport modeling application of the Weather Research and Forecasting model with Chemistry (WRF-Chem). The albedo method estimates sheltering effects on surface wind friction speeds and dust entrainment from the shadows cast by subgrid-scale roughness elements. For this study, we applied the albedo-derived drag partition to the Air Force Weather Agency (AFWA) dust emission module and conducted a sensitivity study on simulated PM10 concentrations using the Georgia Institute of Technology–Goddard Global Ozone Chemistry Aerosol Radiation and Transport (GOCART) model as implemented in WRF-Chem v4.1. Our analysis focused on a convective dust event case study from 3–4 July 2014 for the southwestern United States desert region discussed by other published works. Previous studies have found that WRF-Chem simulations grossly overestimated the dust transport associated with this event. Our results show that removing the default erodibility map and adding the drag parameterization to the AFWA dust module markedly improved the overall magnitude and spatial pattern of simulated dust conditions for this event. Simulated PM10 values near the leading edge of the storm substantially decreased in magnitude (e.g., maximum PM10 values were reduced from 17 151 to 8539 µg m−3), bringing the simulated results into alignment with the observed PM10 measurements. Furthermore, the addition of the drag partition restricted the erroneous widespread dust emission of the original model configuration. We also show that similar model improvements can be achieved by replacing the wind friction speed parameter in the original dust emission module with globally scaled surface wind speeds, suggesting that a well-tuned constant could be used as a substitute for the albedo-based product for short-duration simulations in which surface roughness is not expected to change and for landscapes wherein roughness is constant over years to months. Though this alternative scaling method requires less processing, knowing how to best tune the model winds a priori could be a considerable challenge. Overall, our results demonstrate how dust transport simulation and forecasting with the AFWA dust module can be improved in vegetated drylands by calculating the dust emission flux with surface wind friction speed from a drag partition treatment.

A progressive surrogate gradient learning for memristive spiking neural network

In recent years, spiking neural networks (SNNs) have received increasing attention of research in the field of artificial intelligence due to their high biological plausibility, low energy consumption, and abundant spatio-temporal information. However, the non-differential spike activity makes SNNs more difficult to train in supervised training. Most existing methods focusing on introducing an approximated derivative to replace it, while they are often based on static surrogate functions. In this paper, we propose a progressive surrogate gradient learning for backpropagation of SNNs, which is able to approximate the step function gradually and to reduce information loss. Furthermore, memristor cross arrays are used for speeding up calculation and reducing system energy consumption for their hardware advantage. The proposed algorithm is evaluated on both static and neuromorphic datasets using fully connected and convolutional network architecture, and the experimental results indicate that our approach has a high performance compared with previous research.

Dynam-SLAM: An Accurate, Robust Stereo Visual-Inertial SLAM Method in Dynamic Environments

Most existing vision-based simultaneous localization and mapping (SLAM) systems and their variants still assume that the observation is absolutely static and cannot work well in dynamic environments. Here, we present the Dynam-SLAM (Dynam), a stereo visual-inertial SLAM system capable of robust, accurate, and continuous work in high dynamic environments. Our approach is devoted to loosely coupling the stereo scene flow with an inertial measurement unit (IMU) for dynamic feature detection and tightly coupling the dynamic and static features with the IMU measurements for nonlinear optimization. First, the scene flow uncertainty caused by measurement noise is modeled to derive the accurate motion likelihood of landmarks. Meanwhile, to cope with highly dynamic environments, we additionally construct the virtual landmarks based on the detected dynamic features. Then, we build a tightly coupled, nonlinear optimization-based SLAM system to estimate the camera state by fusing IMU measurements and feature observations. Finally, we evaluate the proposed dynamic feature detection module (DFM) and the overall SLAM system in various benchmark datasets. Experimental results show that the Dynam is almost unaffected by DFM and performs well in static EuRoC datasets. Dynam outperforms the current state-of-the-art visual and visual-inertial SLAM implementations in terms of accuracy and robustness in self-collected dynamic datasets. The average absolute trajectory error of Dynam in the dynamic benchmark datasets is <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"><tex-math notation="LaTeX">$\sim$</tex-math></inline-formula> 90% lower than that of VINS-Fusion, <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"><tex-math notation="LaTeX">$\sim$</tex-math></inline-formula> 84% lower than that of ORB-SLAM3, and <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"><tex-math notation="LaTeX">$\sim$</tex-math></inline-formula> 88% lower than that of Kimera.

TKIFRPM: A Novel Approach for Topmost-K Identical Frequent Regular Patterns Mining from Incremental Datasets

The regular frequent pattern mining (RFPM) approaches are aimed to discover the itemsets with significant frequency and regular occurrence behavior in a dataset. However, these approaches mainly suffer from the following two issues: (1) setting the frequency threshold parameter for the discovery of regular frequent patterns technique is not an easy task because of its dependency on the characteristics of a dataset, and (2) RFPM approaches are designed to mine patterns from the static datasets and are not able to mine dynamic datasets. This paper aims to solve these two issues by proposing a novel top-K identical frequent regular patterns mining (TKIFRPM) approach to function on online datasets. The TKIFRPM maintains a novel synopsis data structure with item support index tables (ISI-tables) to keep summarized information about online committed transactions and dataset updates. The mining operation can discover top-K regular frequent patterns from online data stored in the ISI-tables. The TKIFRPM explores the search space in recursive depth-first order and applies a novel progressive node’s sub-tree pruning strategy to rapidly eliminate a complete infrequent sub-tree from the search space. The TKIFRPM is compared with the MTKPP approach, and it found that it outperforms its counterpart in terms of runtime and memory usage to produce designated topmost-K frequent regular pattern mining on the datasets following incremental updates.

Ensemble Methods for Spatial Data Stream Classification

In this paper, we study the problem of classification of data in a spatial data stream. The classification of spatial objects in a static data set has been well studied. However, classifying objects in a spatial data stream has received very little attention. We propose an iterative ensemble approach for deep learning of a spatial data stream. Using several deep neural networks, our strategy iteratively performs training and testing of the classifier, with the goal of reaching a desired accuracy, and the same accuracy that would be achieved as a classifier that is trained and tested with the entire dataset. An experimental evaluation and comparison of the ensemble approach shows improvement over a previously proposed iterative deep learning strategy.

Upgraded Very Fast Decision Tree: Energy Conservative Algorithm for Data Stream Classification

Traditional machine learning (ML) techniques model knowledge using static datasets. With the increased use of the Internet in today's digital world, a massive amount of data is generated at an accelerated rate that must be handled. This data must be handled as soon as it arrives because it is continuous, and cannot be kept for a long period of time. Various methods exist for mining data from streams. When developing methods like these, the machine learning community put accuracy and execution time first. Numerous sorts of studies take energy consumption into consideration while evaluating data mining methods. However, this work concentrates on Very Fast Decision Tree, which is the most often used technique in data flow classification, despite the fact that it wastes a huge amount of energy on trivial calculations. The research presents a proposed mechanism for upgrading the algorithm's energy usage and restricts computational resources, without compromising the algorithm's efficiency. The mechanism has two stages: the first is to eliminate a set of bad features that increase computational complexity and waste energy, and the second is to group the good features into a candidate group that will be used instead of using all of the attributes in the next iteration. Experiments were conducted on real-world benchmark and synthetic datasets to compare the proposed method to state-of-the-art algorithms in previous works. The proposed algorithm works considerably better and faster with less energy while maintaining accuracy.

Handling concept drift in deep learning applications for process monitoring

As machine learning applications transcend from laboratories to real-world operations in manufacturing use cases such as intelligent maintenance and quality control, questions regarding their continuous reliability and robustness arise. Static datasets used to develop machine learning models can only capture a subset of possible real-world conditions. Instances of concept drift, such as changing environmental-, equipment- and operating conditions may, over time, significantly degrade the performance of machine learning models, compromising safety, acceptance, and economics if not properly addressed. It is thus necessary to (1) detect drifts and (2) efficiently adapt the model to dynamically changing conditions. In this work, we propose a framework to tackle the issue of drift detection by utilizing the underlying neural network's uncertainty. We demonstrate the framework's effectiveness in a case study involving process monitoring of a real-world CNC milling process. The dataset includes accelerometer data and multiple types of realistic concept drifts. In addition, we introduce a methodology for inducing controllable, synthetic concept drift through a simulated sensor rotation. Our experiments show that the introduced drifts lead to a strong performance degradation of the model, which highlights the relevance for practitioners. All drifts are detected using the proposed framework, enabling continuously reliable applications.

Compensation of apparent strain data due to temperature gradients in a full-scale static test of a composite tidal turbine blade

Composite blades play a crucial role in a tidal turbine’s performance, and rigorous testing is required for certification before mass production. This study proposes a methodology for compensating strain measurements affected by temperature gradients using multivariate statistical regressions. This research utilises data analysis and signal processing techniques on strain and temperature datasets obtained from full-scale blade tests at FastBlade, the world???s first regenerative fatigue testing facility for tidal turbine blades. The multivariate regression model contains derived coefficients that separate temperature-induced strains from mechanical strains based on strain gauges coupled to their nearest thermocouples available. The linear model’s performance is assessed against static test datasets with different temperature conditions and scenarios. The proposed compensation method accurately isolated mechanical strains from high-temperature fluctuations from different sources acting on a large structure, enhancing the reliability of full-scale onshore tidal turbine blade testing.

Integrated thermo-poro-mechanical characterization for CO2 Sequestration at deep aquifer conditions

Modelling and forecasting of injected CO2 plume behaviour is an essential step in the baseline, monitoring, and verification [BMV] process in the CO2 sequestration lifecycle. The goal of reduction of uncertainty through forecasting models, can be better realized by accounting for the thermo-poro-mechanical nature of the deep subsurface reservoir systems. The current study focusses on developing and refining a laboratory workflow which will help in generating representative static and dynamic datasets at ambient and deep aquifer conditions. The workflow involves characterizing the poroelastic Biot coefficient and mechanical properties at ambient, high temperatures and at reservoir representative stress conditions. This information will be combined with the dataset from a CO2 flood experiment which replicates the displacement of brine by super critical CO2 at ambient and high temperatures and at reservoir representative stresses. Resistivity and acoustic signals will be monitored throughout the flood experiment. Existing analytical models for fluid substitution such as the Biot-Gassmann-Brie populated with representative data will be evaluated for finding the best description of the experimental observations. The integrated results of the workflow are meant to help develop better informed static and dynamic models improving the confidence in the BMV process of CCUS.

A Spontaneous Driver Emotion Facial Expression (DEFE) Dataset for Intelligent Vehicles: Emotions Triggered by Video-Audio Clips in Driving Scenarios

In this article, a new dataset, the driver emotion facial expression (DEFE) dataset for drivers’ spontaneous emotions analysis is introduced. The dataset includes facial expression recordings from 60 participants during driving. After watching a selected video-audio clip to elicit a specific emotion, each participant completed the driving tasks in the same driving scenario and rated his/her emotional responses during the driving processes from the aspects of dimensional emotion method and discrete emotion method. The study also conducted classification experiments to recognize the scales of arousal, valence, dominance, as well as the emotion category and intensity to establish baseline results for the proposed dataset. Furthermore, this paper compared emotion recognition results difference through facial expressions between dynamic driving and static life scenarios. The results showed that dynamic driving and static life datasets were different in emotion recognition results. To further explore the reasons for the difference in emotion recognition results, the analysis from the AU (action unit) presence perspective was studied. The results showed significant differences in the AUs presence of facial expressions between dynamic driving and static life scenarios, indicating that drivers’ facial expressions may be affected by the driving task to influence the recognition of drivers’ emotions through facial expressions. Therefore, to accurately recognize the drivers’ emotions to establish a reliable emotion-aware human-machine interaction system, thereby improving driving safety and comfort, publishing a human emotion dataset specifically for the driver is necessary. The proposed dataset will be publicly available so that researchers worldwide can use it to develop and examine their driver emotion analysis methods. To the best of our knowledge, this is currently the only public driver facial expression dataset.

Recurrence quantification analysis of center of pressure trajectories for balance and fall-risk assessment in young and older adults.

The prevalence and impact of balance impairments and falls in older adults have motivated several studies on the characterization of human balance. This study aimed to determine the ability of recurrence quantification analysis (RQA) measures to characterize balance control during quiet standing in young and older adults and to discriminate between different fall risk groups. We analyze center pressure trajectories in the medial-lateral and anterior-posterior directions from a publicly available static posturography dataset that contains tests acquired under four vision-surface testing conditions. Participants were retrospectively classified as young adults (age<60, n=85), non-fallers (age≥60, falls=0, n=56), and fallers (age≥60, falls≥1, n=18). Mixed ANOVA and post hoc analyzes were performed to test for differences between groups. For CoP fluctuations in the anterior-posterior direction, all RQA measures showed significantly higher values for young than older adults when standing on a compliant surface, indicating less predictable and stable balance control among seniors under testing conditions where sensory information is restricted or altered. However, no significant differences between non-fallers and fallers were observed. These results support the use of RQA to characterize balance control in young and old adults, but not to discriminate between different fall risk groups.

Video logo detection by Deep-Transfer Active Learning

Brand logo detection is a special aspect of machine vision. However, Video logo detection benchmarks are scarce in the public domain. We exploit the power of a deep convolutional neural network (DCNN) and leverage established datasets related to existing applications to develop a deep-transfer active-learning (DTAL) algorithm to select the most valuable samples so that the smallest number possible needs to be labeled to achieve maximum performance improvements for video object detection model training. By exploiting the possible shared deep feature space between static and video datasets through transfer learning based on highly adaptable DCNN features, DTAL implements diversity-based active learning to select the most informative samples from a sequence of similar image frames for video object detection. We successfully apply the new DTAL algorithm to implement active learning for logo detection from live streaming sports videos as well as pedestrian and face detection from video data. We show that DTAL is a better active-learning method than state-of-the-art deep-learning-based active-learning object detection techniques. We also contribute one of the largest video-based logo resources, the Sports Match Video Logo (SMVL) dataset, to facilitate general logo detection research using transfer- and active-learning applications for video object detection.

Efficient and Privacy-Preserving Arbitrary Polygon Range Query Scheme Over Dynamic and Time-Series Location Data

Location-based services (LBSs) provide enhanced functionality of mobile applications and convenience for mobile users, which plays a more and more remarkable role in people’s daily life. In LBSs, spatial range query is an essential tool for users to find interesting points in a specific region. However, during spatial range query, it is necessary for data owners and query users to exchange their location data, and the leakage of private location information has drawn significant attention in both governmental and social aspects. Meanwhile, most existing location privacy protection schemes only focus on achieving regular geometry range query over static location datasets. In this paper, we present an efficient and privacy-preserving arbitrary polygon range query scheme, named EPAPRQ. With EPAPRQ, the arbitrary and fine-grained polygon range query can be executed over a dynamic and time-series location dataset with privacy protection. Specifically, in EPAPRQ, an arbitrary polygon range query algorithm is first introduced with sub-range query technique. Then, to protect the private location information of the data owner and query users, a series privacy-preserving data computation protocols are constructed with a symmetric homomorphic encryption algorithm, and a ciphertext-based location dataset updating strategy is also designed. Finally, we propose a double filtration method through combining the quadtree index structure and minimum bounded rectangle, which is able to greatly improve the query efficiency over ciphertexts. Detailed security analysis shows that the sensitive location information in EPAPRQ can be well protected. Furthermore, we evaluate the performance of EPAPRQ in the real map, and the results demonstrate that EPAPRQ is indeed efficient.