Stream Machine Research Articles

AdaDeepStream: streaming adaptation to concept evolution in deep neural networks

Typically, Deep Neural Networks (DNNs) are not responsive to changing data. Novel classes will be incorrectly labelled as a class on which the network was previously trained to recognise. Ideally, a DNN would be able to detect changing data and adapt rapidly with minimal true-labelled samples and without catastrophically forgetting previous classes. In the Online Class Incremental (OCI) field, research focuses on remembering all previously known classes. However, real-world systems are dynamic, and it is not essential to recall all classes forever. The Concept Evolution field studies the emergence of novel classes within a data stream. This paper aims to bring together these fields by analysing OCI Convolutional Neural Network (CNN) adaptation systems in a concept evolution setting by applying novel classes in patterns. Our system, termed AdaDeepStream, offers a dynamic concept evolution detection and CNN adaptation system using minimal true-labelled samples. We apply activations from within the CNN to fast streaming machine learning techniques. We compare two activation reduction techniques. We conduct a comprehensive experimental study and compare our novel adaptation method with four other state-of-the-art CNN adaptation methods. Our entire system is also compared to two other novel class detection and CNN adaptation methods. The results of the experiments are analysed based on accuracy, speed of inference and speed of adaptation. On accuracy, AdaDeepStream outperforms the next best adaptation method by 27% and the next best combined novel class detection/CNN adaptation method by 24%. On speed, AdaDeepStream is among the fastest to process instances and adapt.

Using Stream Data Processing for Real-Time Occupancy Detection in Smart Buildings.

Controlling active and passive systems in buildings with the aim of optimizing energy efficiency and maintaining occupants’ comfort is the major task of building management systems. However, most of these systems use a predefined configuration, which usually do not match the occupants’ preferences. Therefore, occupancy detection is imperative for energy use management mainly in residential and industrial buildings. Most works related to data-driven-based occupancy detection have used batch learning techniques, which need to store first and then train the data. It is not appropriate for a non-stationary environment. Therefore, this work sheds more light on the use of non-stationary machine learning techniques. To this end, three machine learning algorithms for stream data processing are presented, tested, and evaluated in term of accuracy and resources performance (i.e., RAM, CPU), with the aim of predicting the number of occupants in smart buildings. A platform architecture that integrates IoT technologies with stream machine learning is implemented and deployed. The experimental results show the effectiveness of this approach and illustrate that the number of occupants can be predicted with an accuracy of more than 83% and without resource wasting (i.e., time of CPU use varied between 0.04s and 3.85 ⋅ 10−11 GB of RAM could be exploited per hour).

An extensive study of C-SMOTE, a Continuous Synthetic Minority Oversampling Technique for Evolving Data Streams

Streaming Machine Learning (SML) studies algorithms that update their models, given an unbounded and often non-stationary flow of data performing a single pass. Online class imbalance learning is a branch of SML that combines the challenges of both class imbalance and concept drift. In this paper, we investigate the binary classification problem by rebalancing an imbalanced stream of data in the presence of concept drift, accessing one sample at a time. We propose an extensive comparative study of Continuous Synthetic Minority Oversampling Technique (C-SMOTE), inspired by the popular sampling technique Smote, as a meta-strategy to pipeline with SML classification algorithms. We benchmark C-SMOTE pipelines on both synthetic and real data streams, containing different types of concept drifts, different imbalance levels, and different class distributions. We bring statistical evidence that models learnt with C-SMOTE pipelines improve the minority class performance concerning both the baseline models and the state-of-the-art methods. We also perform a sensitivity analysis to detect the C-SMOTE impact on the majority class performance for the three types of concept drift and several class distributions. Moreover, we show a computational cost analysis in terms of time and memory consumption.

Enabling Low-Redundancy Proactive Fault Tolerance for Stream Machine Learning Via Erasure Coding: Design and Evaluation

Enabling Low-Redundancy Proactive Fault Tolerance for Stream Machine Learning Via Erasure Coding: Design and Evaluation

Predict COVID-19 Spreading With C-SMOTE

Data continuously gathered monitoring the spreading of the COVID-19 pandemic form an unbounded flow of data. Accurately forecasting if the infections will increase or decrease has a high impact, but it is challenging because the pandemic spreads and contracts periodically. Technically, the flow of data is said to be imbalanced and subject to concept drifts because signs of decrements are the minority class during the spreading periods, while they become the majority class in the contraction periods and the other way round. In this paper, we propose a case study applying the Continuous Synthetic Minority Oversampling Technique (C-SMOTE), a novel meta-strategy to pipeline with Streaming Machine Learning (SML) classification algorithms, to forecast the COVID-19 pandemic trend. Benchmarking SML pipelinesthat use C-SMOTE against state-of-the-art methods on a COVID-19 dataset, we bring statistical evidence that models learned using C-SMOTE are better.

Towards Strong AI

Strong AI—artificial intelligence that is in all respects at least as intelligent as humans—is still out of reach. Current AI lacks common sense, that is, it is not able to infer, understand, or explain the hidden processes, forces, and causes behind data. Main stream machine learning research on deep artificial neural networks (ANNs) may even be characterized as being behavioristic. In contrast, various sources of evidence from cognitive science suggest that human brains engage in the active development of compositional generative predictive models (CGPMs) from their self-generated sensorimotor experiences. Guided by evolutionarily-shaped inductive learning and information processing biases, they exhibit the tendency to organize the gathered experiences into event-predictive encodings. Meanwhile, they infer and optimize behavior and attention by means of both epistemic- and homeostasis-oriented drives. I argue that AI research should set a stronger focus on learning CGPMs of the hidden causes that lead to the registered observations. Endowed with suitable information-processing biases, AI may develop that will be able to explain the reality it is confronted with, reason about it, and find adaptive solutions, making it Strong AI. Seeing that such Strong AI can be equipped with a mental capacity and computational resources that exceed those of humans, the resulting system may have the potential to guide our knowledge, technology, and policies into sustainable directions. Clearly, though, Strong AI may also be used to manipulate us even more. Thus, it will be on us to put good, far-reaching and long-term, homeostasis-oriented purpose into these machines.

Improved Real-Time Natural Hazard Monitoring Using Automated DInSAR Time Series

As part of the collaborative GeoSciFramework project, we are establising a monitoring system for the Yellowstone volcanic area that integrates multiple geodetic and seismic data sets into an advanced cyber-infrastructure framework that will enable real-time streaming data analytics and machine learning and allow us to better characterize associated long- and short-term hazards. The goal is to continuously ingest both remote sensing (GNSS, DInSAR) and ground-based (seismic, thermal and gas observations, strainmeter, tiltmeter and gravity measurements) data and query and analyse them in near-real time. In this study, we focus on DInSAR data processing and the effects from using various atmospheric corrections and real-time orbits on the automated processing and results. We find that the atmospheric correction provided by the European Centre for Medium-Range Weather Forecasts (ECMWF) is currently the most optimal for automated DInSAR processing and that the use of real-time orbits is sufficient for the early-warning application in question. We show analysis of atmospheric corrections and using real-time orbits in a test case over the Kilauea volcanic area in Hawaii. Finally, using these findings, we present results of displacement time series in the Yellowstone area between May 2018 and October 2019, which are in good agreement with GNSS data where available. These results will contribute to a baseline model that will be the basis of a future early-warning system that will be continuously updated with new DInSAR data acquisitions.

A hybrid distributed batch-stream processing approach for anomaly detection

Batch and stream processing are separately and efficiently applied in many applications. However, some newer data-driven applications such as the Internet of Things and cloud computing call for hybrid processing approaches in order to handle the speed and accuracy required for processing such complex data. In this paper, we propose a Hybrid Distributed Batch-Stream (HDBS) architecture for anomaly detection in real-time data. The hybrid architecture, while benefiting from the accuracy provided by batch processing, also enjoys the speed and real-time features of stream processing. In the proposed architecture, our focus is on the algorithmic aspects of hybrid processing including the interaction models between batch and stream processing units, the characteristics of batch and stream machine learning algorithms and the principles of merging the results of different processing units. The driving idea of such combination is that the results of batch and stream processing units are complementary with each other, as one of them constructs accurate models based on previous data, and the other one is capable of processing new stream data in real-time. Furthermore, we propose a generalized version of the HDBS with respect to its algorithms and communication policy levels. In the generalized HDBS architecture, we address the various aspects of the interaction between the batch and stream processing units, and the merging operations to produce the final results. the evaluations of the proposed architecture using various criteria (accuracy, space complexity, and time complexity) demonstrate that the accuracy of the proposed method is higher than the accuracy of the batch processing methods, its time complexity is also similar to one of the stream processing methods and much less than the batch processing methods, which makes our proposed architecture an efficient and practical solution for real-time anomaly detection.

Real-Time Healthcare Monitoring System using Online Machine Learning and Spark Streaming

The real-time monitoring and tracking systems play a critical role in the healthcare field. Wearable medical devices with sensors, mobile applications, and health cloud have continuously generated an enormous amount of data, often called streaming big data. Due to the higher speed of the streaming data, it is difficult to ingest, process, and analyze such huge data in real-time to make real-time actions in case of emergencies. Using traditional methods that are inadequate and time-consuming. Therefore, there is a significant need for real-time big data stream processing to guarantee an effective and scalable solution. So, we proposed a new system for online prediction to predict health status using Spark streaming framework. The proposed system focuses on applying streaming machine learning models (i.e. streaming linear regression with SGD) on streaming health data events ingested to spark streaming through Kafka topics. The experimental results are done on the historical medical datasets (i.e. diabetes dataset, heart disease dataset, and breast cancer dataset) and generated dataset which is simulated to wearable medical sensors. The historical datasets have shown that the accuracy improvement ratio obtained using the diabetes disease dataset is the highest one with respect to the other two datasets with an accuracy of 81%. For generated datasets, the online prediction system has achieved accuracy with 98% at 5 seconds window size. Beyond this, the experimental results have proofed that the online prediction system can online learn and update the model according to the new data arrival and window size.

Effect of Feed Rate on Surface Roughness in Abrasive Waterjet Machining of Jute Fibre Reinforced Polyester Composites

The common fiber strengthened composite are as a rule progressively utilized in different applications like space, flying machine, marine, building and car segment on account of their prevalent physical and mechanical properties despite the fact that they are somewhat expensive. AWJM speaks to a worldwide adaptable instrument supporting the machining of every normal material that are with no harm by straight contact with water. In this present work plans to explore jute fiber strengthened unsaturated polyester resin composite material investigating surface Roughness by changing three parameters, for example, Pressure, Standoff distance and Feed rate an ideal state of the material can be found and furthermore investigation of the cut surface geography after grating water stream machining. Rough utilized for cutting is garnet 80 mesh abrasive particles. It catalyzes the cutting and gives phenomenal surface completion.

Estimation and Comparison of Electrode Wear and Ae Parameters of Titanium Material in Wire Electric Discharge Machining Using ANN

Wire Electrical Discharge Machining (WEDM) is a specialized thermal machining process capable of accurately machining parts with varying hardness or complex shapes, which have sharp edges that are very difficult to be machined by the main stream machining processes. Selection of process parameters for obtaining higher cutting efficiency or accuracy in WEDM is still not fully solved, even with most up-to-date CNC wire EDM machine. It is widely recognised that Acoustic Emission (AE) is gaining ground as a monitoring method for health diagnosis on rotating machinery. The advantage of AE monitoring over vibration monitoring is that the AE monitoring can detect the growth of subsurface cracks whereas the vibration monitoring can detect defects only when they appear on the surface. This study outlines the optimization of titanium material using L16design of experiment. Each experiment has been performed varying the process parameters like pulse-on time, pulse-off time, current and bed speed. Among different process parameters voltage and flush rate were kept constant. Molybdenum wire having diameter of 0.18 mm was used as an electrode. Simple functional relationships between the parameters were plotted to arrive at possible information on Electrode Wear (EW) and AE signals. But these simpler methods of analysis did not provide any information about the status of the electrode. Thus, there is a requirement for more sophisticated methods that are capable of integrating information from the multiple sensors. Hence, method like Artificial Neural Network (ANN) has been applied for the estimation of EW, AE signal strength, AE count and AE RMS. The ANN algorithm is designed to learn the process by training the algorithm with the experimental data. The experimental observations are divided into three sets: the training set, validation set and testing set. The training set is used to make the ANN learn the process and the testing set will check the performance of ANN. Different models can be obtained by varying the percentage of data in the training set and the best model can be selected from these, viz., 50%, 60% and 70%. The best model is selected from the said percentages of data. Estimation of the EW and AE signals parameters by ANN at 70% of data training set showed the best correlation with the measured value.

Multi-granularity bidirectional attention stream machine comprehension method for emotion cause extraction

Emotion cause extraction is to extract the cause information that triggers the emotion expression of described person. This task in text plays a critical role in natural language processing applications, such as sentiment analysis and semantic comprehension system. However, most existing methods for this emotion cause extraction task only focus on feature engineering and ignore the latent semantic information between emotion word and context to hinder the performance. In this paper, we propose a novel computational multi-granularity bidirectional attention stream network based on a machine comprehension frame to settle this problem. The context and query are embedded by this multistage hierarchical process based on the fine-grained levels of embeddings. Then, the bidirectional attention stream mechanism is applied to get an emotional query-aware context representation. Meanwhile, we have conducted extensive experiments on available Chinese emotion cause dataset. The experimental results demonstrate that our approach significantly outperforms the state-of-the-art methods and is able to extract the emotion cause.

A Shared Memory Parallel Block Streaming Model for Irregular Applications

Due to worsening machine balance, a lightweight irregular application can utilize only a small fraction of the peak computational capacity on modern processors. Performance of such an application is also unpredictable due to the scattered data accesses. Even though architectural features such as a cache system, hardware prefetchers etc., that reduce the cost of irregular access, are commonly found in most of the modern processors, their design parameters differ widely from one processor to another. Therefore, a performance improving programming technique still needs extensive tuning to gain maximum benefit on a target processor. In this scenario, achieving portable performance becomes difficult. This work proposes a block streaming machine model and hypothesizes that an algorithm based on the model has predictable execution time. To enable adaptation of this model for irregular applications, we also provide algorithmic transformations that can be used to replace the scattered accesses with streaming accesses in a cost predictable way. Further, we experimentally demonstrate usefulness of the model and the transformations for static lightweight irregular computations such as those performed by a numerical partial differential equation solver on modern multicore processors.

On-line Elastic Similarity Measures for time series

The way similarity is measured among time series is of paramount importance in many data mining and machine learning tasks. For instance, Elastic Similarity Measures are widely used to determine whether two time series are similar to each other. Indeed, in off-line time series mining, these measures have been shown to be very effective due to their ability to handle time distortions and mitigate their effect on the resulting distance. In the on-line setting, where available data increase continuously over time and not necessary in a stationary manner, stream mining approaches are required to be fast with limited memory consumption and capable of adapting to different stationary intervals. In this sense, the computational complexity of Elastic Similarity Measures and their lack of flexibility to accommodate different stationary intervals, make these similarity measures incompatible with the requirements mentioned. To overcome these issues, this paper adapts the family of Elastic Similarity Measures – which includes Dynamic Time Warping, Edit Distance, Edit Distance for Real Sequences and Edit Distance with Real Penalty – to the on-line setting. The proposed adaptation is based on two main ideas: a forgetting mechanism and the incremental computation. The former makes the similarity consistent with streaming time series characteristics by giving more importance to recent observations, whereas the latter reduces the computational complexity by avoiding unnecessary computations. In order to assess the behavior of the proposed similarity measure in on-line settings, two different experiments have been carried out. The first aims at showing the efficiency of the proposed adaptation, to do so we calculate and compare the computation time for the elastic measures and their on-line adaptation. By analyzing the results drawn from a distance-based streaming machine learning model, the second experiment intends to show the effect of the forgetting mechanism on the resulting similarity value. The experimentation shows, for the aforementioned Elastic Similarity Measures, that the proposed adaptation meets the memory, computational complexity and flexibility constraints imposed by streaming data.

Prediction method for creep life of thin-wall specimen with film cooling holes in Ni-based single-crystal superalloy

Considering the effects of numbers and drill processes of film cooling holes, the effect of a multiaxial stress state on the creep behavior of Ni-based single-crystal superalloy was investigated. The thin-wall plate specimens with film cooling holes, including multiholes and a single central hole, have been designed. The drilling processes are electro stream machining (ESM), electrical discharge machining (EDM) and laser drilling (LD). The results show that creep lives were longer in the thin-wall specimens with a single central hole and shorter in specimens with multi-holes due to the multihole interference effect. Under the same stress and temperature, the specimens with multiholes using the LD process have the longest creep lives compared with the other two processes. To predict the creep life of a specimen with film cooling holes, a creep damage constitutive model under the crystal plasticity framework was proposed by adding geometric structure variables β and m. Coupling with the modified skeletal point stress method (SPSM), the creep rupture lives of the specimens with film cooling holes were well predicted. The error of creep life between the finite element (FE) calculation and experimental results is within 28.03%.

Analysis of lean manufacturing in an automobile industry - a case study

The automobile industries are more competitive and customer demand is volatile. The customers are willing to buy the product at low cost with high quality. So common principle is created to reduce the wastage and increase the efficiency of the industry. Lean manufacturing is the best principle to adopt the manufacturing industry. The main aim of this project is reduce the wastage, cost and increase the productivity of the industry. Using lean tools just in time (JIT), kanban, value stream mapping (current and future state) and man machine chart (current and future) for reduce wastage and cost. In current state value stream map, the waste area is identified. Finally, non-value added time is reduced to 28.7%. The value added time is reduced to 27.84%. The percentage value addition is increased from 3.25% to 3.29%, the total inventory is reduced to 25.07% and man machine ratio is improved as 1.52: 1.26.

Primena injektora za otsisavanja hemijski agresinih gasova

The injector is a fluid machine of simple construction, featuring no moving parts. It is fluid stream machine. Should there be an occasional need for exhausting chemically aggressive fluids, this machine proves very convenient on account of its lower purchase price compared to other types of hydraulic machines. The injector discussed in this article has been originally designed for exhausting gases from the sulphur chamber, which is used in the pretreatment of fruit before drying. Once the sulphuring is completed, the SO2 gas should be exhausted promptly. Under realistic sulphuring conditions, the effect measurement was performed at a compressed air pressure of 2 bar. The compressed air pressures ranged from 0.5 to 2.5 bar. In the instance when the chamber was filled with fruit, at a pressure of 2 bar, the ratio between the exhausted and compressed air was 1.68, at an exhausted gas flow rate of 0.0327 kg/s. The volume of empty chamber was 0.2856 m3. According to these data, a period of 10.04 s is required for one air change in the chamber. As complete aeration of the empty chamber necessitates three air changes, the injector should be in operation for 30.12 s.

Study of Surface Roughness and AE Signals while Machining Titanium Grade-2 Material using ANN in WEDM

Wire Electrical Discharge Machining (WEDM) is a specialized thermal machining process capable of accurately machining parts with varying hardness or complex shapes, which have sharp edges that are very difficult to be machined by the main stream machining processes. It is widely recognized that Acoustic Emission (AE) is gaining ground as a monitoring method for health diagnosis on rotating machinery. The advantage of AE monitoring over vibration monitoring is that the AE monitoring can detect the growth of subsurface cracks whereas the vibration monitoring can detect defects only when they appear on the surface. This study outlines the development of model and its application to estimation of machining performances using Artificial Neural Network (ANN). Each experiment has been performed under different process parameters of pulse-on time, pulse-off time, current and bed speed. Among different process parameters voltage and flush rate were kept constant. Molybdenum wire having diameter of 0.18mm was used as an electrode. Estimation and comparison of responses like surface roughness and AE signals was carried out using ANN.

Comparison of Machining Responses Using Multiple Regression Analysis and Group Method Data Handling Technique of EN-19 Material in WEDM

Wire Electrical Discharge Machining (WEDM) is a specialized thermal machining process capable of accurately machining parts with varying hardness or complex shapes, which have sharp edges that are very difficult to be machined by the main stream machining processes. This study outlines the development of model and its application to estimation and comparison of machining responses using Multiple Regression Analysis (MRA) and Group Method Data Handling Technique (GMDH). Experimentation was performed as per Taguchi’s L’16 orthogonal array for EN-19 material. Each experiment has been performed under different cutting conditions of pulse-on, pulse-off, current and bed speed. Among different process parameters voltage and flush rate were kept constant. Molybdenum wire having diameter of 0.18 mm was used as an electrode. Four responses namely accuracy, surface roughness, volumetric material removal rate and electrode wear have been considered for each experiment. Estimation and comparison of responses was carried out using MRA and GMDH.

Comparative Study of Electrode Wear Estimation in Wire EDM using Multiple Regression Analysis and Group Method Data Handling Technique for EN-8 and EN-19

Wire Electrical Discharge Machining (WEDM) is a specialized thermal machining process capable of accurately machining parts with varying hardness or complex shapes, which have sharp edges that are very difficult to be machined by the main stream machining processes. In WEDM a specific wire run-off speed is applied to compensate wear and avoid wire breakage. Since the workpiece generally stays stationary and short discharge durations are applied, the relative displacement between wire and workpiece during one single discharge is very small. This study outlines the development of model and its application to optimize WEDM machining parameters using the Taguchi's technique which is based on the robust design. Present study outlines the electrode wear estimation in the wire EDM. EN-8 and EN-19 was machined using different process parameters based on L'16 orthogonal array. Among different process parameters voltage and flush rate were kept constant. Parameters such as bed speed, current, pulse-on and pulse-off was varied. Molybdenum wire having diameter of 0.18 mm was used as an electrode. Electrode wear was measured using universal measuring machine. Estimation and comparison of electrode wear was done using multiple regression analysis and group method data handling technique. From the results it was observed that, measured electrode wear and estimated electrode wear correlates well with respect to MRA than GMDH.