Preparation Of Training Data Research Articles

Data augmentation in microscopic images for material data mining

Recent progress in material data mining has been driven by high-capacity models trained on large datasets. However, collecting experimental data (real data) has been extremely costly owing to the amount of human effort and expertise required. Here, we develop a novel transfer learning strategy to address problems of small or insufficient data. This strategy realizes the fusion of real and simulated data and the augmentation of training data in a data mining procedure. For a specific task of grain instance image segmentation, this strategy aims to generate synthetic data by fusing the images obtained from simulating the physical mechanism of grain formation and the “image style” information in real images. The results show that the model trained with the acquired synthetic data and only 35% of the real data can already achieve competitive segmentation performance of a model trained on all of the real data. Because the time required to perform grain simulation and to generate synthetic data are almost negligible as compared to the effort for obtaining real data, our proposed strategy is able to exploit the strong prediction power of deep learning without significantly increasing the experimental burden of training data preparation.

SAFETY RIDING, BENTUK PELAKSANAAN PELATIHAN PADA PT SARANA AMAN BERKENDARA JAKARTA SELATAN

<p>This study aims to find out how the form of conducting safety riding training at PT Sarana Aman Berkendara, find out the constraints faced and find out the solutions made in overcoming the obstacles faced.The form of implementation and training in safety rite consists of 3 (three) stages, namely; pre test, theory & practice material, Post test. In the implementation of safety riding training there are several obstacles, namely Lack of order among the trainees so that it is not in accordance with the time provided, Lack of preparation of training participants' data, Limited office facilities, Weather that often changes, Number of participants. Solutions in dealing with obstacles in the implementation of safety riding training that is Handling obstacles in the lack of order of the trainees, Handling obstacles in the lack of preparation of participants' training data, Handling constraints in limited office facilities, Handling constraints on the weather that often changes, Handling constraints in the number of participants. This research is descriptive qualitative, with observation, interview and documentation techniques.</p><p><strong>Keywords</strong>: Safety Riding, Implementation, Training</p>

Deep learning-enhanced fluorescence microscopy via degeneration decoupling.

Deep learning-based reconstruction has emerged as an effective tool in fluorescence microscopy, with the potential to resolve diffraction-limited structures. However, most deep-learning reconstruction methods employed an end-to-end strategy, which ignored physical laws in the imaging process and made the preparation of training data highly challenging as well. In this study, we have proposed a novel deconvolution algorithm based on an imaging model, deep-learning priors and the alternating direction method of multipliers. This scheme decouples the reconstruction into two separate sub-problems, one for deblurring and one for restraining noise and artifacts. As a result of the decoupling, we are able to introduce deep-learning image priors and a variance stabilizing transform against targeted image degeneration due to the low photon budget. Deep-learning priors are learned from the general image dataset, in which biological images do not have to be involved, and are more powerful than hand-designed ones. Moreover, the use of the imaging model ensures high fidelity and generalization. Experiments on various kinds of measurement data show that the proposed algorithm outperforms existing state-of-the-art deconvolution algorithms in resolution enhancement and generalization.

Improving Code-mixed POS Tagging Using Code-mixed Embeddings

Social media data has become invaluable component of business analytics. A multitude of nuances of social media text make the job of conventional text analytical tools difficult. Code-mixing of text is a phenomenon prevalent among social media users, wherein words used are borrowed from multiple languages, though written in the commonly understood roman script. All the existing supervised learning methods for tasks such as Parts Of Speech (POS) tagging for code-mixed social media (CMSM) text typically depend on a large amount of training data. Preparation of such large training data is resource-intensive, requiring expertise in multiple languages. Though the preparation of small dataset is possible, the out of vocabulary (OOV) words pose major difficulty, while learning models from CMSM text as the number of different ways of writing non-native words in roman script is huge. POS tagging for code-mixed text is non-trivial, as tagging should deal with syntactic rules of multiple languages. The important research question addressed by this article is whether abundantly available unlabeled data can help in resolving the difficulties posed by code-mixed text for POS tagging. We develop an approach for scraping and building word embeddings for code-mixed text illustrating it for Bengali-English, Hindi-English, and Telugu-English code-mixing scenarios. We used a hierarchical deep recurrent neural network with linear-chain CRF layer on top of it to improve the performance of POS tagging in CMSM text by capturing contextual word features and character-sequence–based information. We prepared a labeled resource for POS tagging of CMSM text by correcting 19% of labels from an existing resource. A detailed analysis of the performance of our approach with varying levels of code-mixing is provided. The results indicate that the F1-score of our approach with custom embeddings is better than the CRF-based baseline by 5.81%, 5.69%, and 6.3% in Bengali, Hindi , and Telugu languages, respectively.

Data-Driven Relay Selection for Physical-Layer Security: A Decision Tree Approach

Conventional optimization-driven secure relay selection relies on maximization algorithm and accurate channel state information (CSI) of both legitimate and eavesdropper channels. Particularly, estimating and collecting accurate eavesdropper CSI is a difficult task. In this paper, we exploit the benefits of machine learning in solving secure relay selection problem from a data-driven perspective. We convert secure relay selection to a multiclass-classification problem and solve it by a decision-tree-based scheme, which is composed of three phases - preparing training data, building decision tree and predicting relay selection. To meet decision tree’s requirement that input features must take discrete values, a feature extraction method is proposed to generate discrete input by quantizing the accurate CSI of legitimate and eavesdropper channels. By this means, the decision-tree-based relay selection only requires quantized CSI feedback which takes substantially fewer bits in predicting phase. For the purpose of optimizing quantization parameters and enhancing decision tree prediction, we further derive three splitting criteria, i.e. information gain, information gain ratio and Gini index. Simulation results show that if the quantization parameters are set properly, the proposed decision-tree-based scheme can achieve satisfactory performance in terms of average secrecy rate while reducing computational complexity and feedback amount.

Deep Neural Network for Estimation of Direction of Arrival With Antenna Array

For many object tracking systems, how to quickly and efficiently estimate the direction of arrival (DOA) of radio waves impinging on the antenna array is an urgent task. In this paper, a new efficient DOA estimation approach based on the deep neural networks (DNN) is proposed, in which a nonlinear mapping that relates the outputs of the receiving antennas with its associated DOAs is learned by using the DNN-based network. The novel network architecture is divided into two phases, the detection phase and the DOA estimation phase. Additional detection network dramatically reduces the size of the training set and the process of the training data preparation is discussed in detail. After finishing the training phase, the corresponding DOAs can be identified based on current input data during testing phase. It has been shown that the proposed method can not only achieve reasonably high DOA estimation accuracy, but also reduce the computational complexity required by traditional superresolution DOA estimation methods such as multiple signal classification (MUSIC) and estimation of signal parameters via rotation invariance (ESPRIT). The computer simulation results are performed to investigate the generalization and effectiveness of the proposed approach in different scenarios.

Efficient and effective OCR engine training

We present an efficient and effective approach to train OCR engines using the Aletheia document analysis system. All components required for training are seamlessly integrated into Aletheia: training data preparation, the OCR engine’s training processes themselves, text recognition, and quantitative evaluation of the trained engine. Such a comprehensive training and evaluation system, guided through a GUI, allows for iterative incremental training to achieve best results. The widely used Tesseract OCR engine is used as a case study to demonstrate the efficiency and effectiveness of the proposed approach. Experimental results are presented validating the training approach with two different historical datasets, representative of recent significant digitisation projects. The impact of different training strategies and training data requirements is presented in detail.

U-Net-bin: hacking the document image binarization contest

Image binarization is still a challenging task in a variety of applications. In particular, Document Image Binarization Contest (DIBCO) is organized regularly to track the state-of-the-art techniques for the historical document binarization. In this work we present a binarization method that was ranked first in the DIBCO`17 contest. It is a convolutional neural network (CNN) based method which uses U-Net architecture, originally designed for biomedical image segmentation. We describe our approach to training data preparation and contest ground truth examination and provide multiple insights on its construction (so called hacking). It led to more accurate historical document binarization problem statement with respect to the challenges one could face in the open access datasets. A docker container with the final network along with all the supplementary data we used in the training process has been published on Github.

A deep learning based pipeline for optical coherence tomography angiography.

Optical coherence tomography angiography (OCTA) is a relatively new imaging modality that generates microvasculature map. Meanwhile, deep learning has been recently attracting considerable attention in image-to-image translation, such as image denoising, super-resolution and prediction. In this paper, we propose a deep learning based pipeline for OCTA. This pipeline consists of three parts: training data preparation, model learning and OCTA predicting using the trained model. To be mentioned, the datasets used in this work were automatically generated by a conventional system setup without any expert labeling. Promising results have been validated by in-vivo animal experiments, which demonstrate that deep learning is able to outperform traditional OCTA methods. The image quality is improved in not only higher signal-to-noise ratio but also better vasculature connectivity by laser speckle eliminating, showing potential in clinical use. Schematic description of the deep learning based optical coherent tomography angiography pipeline.

Urban Tree Species Classification Using a WorldView-2/3 and LiDAR Data Fusion Approach and Deep Learning.

Urban areas feature complex and heterogeneous land covers which create challenging issues for tree species classification. The increased availability of high spatial resolution multispectral satellite imagery and LiDAR datasets combined with the recent evolution of deep learning within remote sensing for object detection and scene classification, provide promising opportunities to map individual tree species with greater accuracy and resolution. However, there are knowledge gaps that are related to the contribution of Worldview-3 SWIR bands, very high resolution PAN band and LiDAR data in detailed tree species mapping. Additionally, contemporary deep learning methods are hampered by lack of training samples and difficulties of preparing training data. The objective of this study was to examine the potential of a novel deep learning method, Dense Convolutional Network (DenseNet), to identify dominant individual tree species in a complex urban environment within a fused image of WorldView-2 VNIR, Worldview-3 SWIR and LiDAR datasets. DenseNet results were compared against two popular machine classifiers in remote sensing image analysis, Random Forest (RF) and Support Vector Machine (SVM). Our results demonstrated that: (1) utilizing a data fusion approach beginning with VNIR and adding SWIR, LiDAR, and panchromatic (PAN) bands increased the overall accuracy of the DenseNet classifier from 75.9% to 76.8%, 81.1% and 82.6%, respectively. (2) DenseNet significantly outperformed RF and SVM for the classification of eight dominant tree species with an overall accuracy of 82.6%, compared to 51.8% and 52% for SVM and RF classifiers, respectively. (3) DenseNet maintained superior performance over RF and SVM classifiers under restricted training sample quantities which is a major limiting factor for deep learning techniques. Overall, the study reveals that DenseNet is more effective for urban tree species classification as it outperforms the popular RF and SVM techniques when working with highly complex image scenes regardless of training sample size.

Extracting the Pairs of Opinion Target and Opinion Term From Reviews With Adaptive Crowd Labeling

Labeling samples manually is a laborious, error-prone, and cost-consuming process, which is a traditional approach to preparing training data for supervised learning. Crowdsourcing provides an effective way to acquire labeled training data. In this paper, we propose an adaptive crowd labeling method to construct a set of the pairs of opinion target and opinion term iteratively under certain budget constraint. First, we assess the workers’ reliabilities with a small number of labeled samples based on an EM process, since the worker’s expertise varies widely on an open crowdsourcing platform. And then, the tasks are assigned to the high reliable workers for labeling the pairs of opinion target and opinion term. Finally, the responds of a sample from multiple workers are integrated to generate a final result based on the labelers’ reliabilities as well as the dependency relation between opinion target and opinion term in this sample. A series of the experimental results show that the proposed method can achieve better extracting effectiveness compared with the baselines and the state-of-the-art methods.

Automatic Mapping of Thermokarst Landforms from Remote Sensing Images Using Deep Learning: A Case Study in the Northeastern Tibetan Plateau

Thawing of ice-rich permafrost causes thermokarst landforms on the ground surface. Obtaining the distribution of thermokarst landforms is a prerequisite for understanding permafrost degradation and carbon exchange at local and regional scales. However, because of their diverse types and characteristics, it is challenging to map thermokarst landforms from remote sensing images. We conducted a case study towards automatically mapping a type of thermokarst landforms (i.e., thermo-erosion gullies) in a local area in the northeastern Tibetan Plateau from high-resolution images by the use of deep learning. In particular, we applied the DeepLab algorithm (based on Convolutional Neural Networks) to a 0.15-m-resolution Digital Orthophoto Map (created using aerial photographs taken by an Unmanned Aerial Vehicle). Here, we document the detailed processing flow with key steps including preparing training data, fine-tuning, inference, and post-processing. Validating against the field measurements and manual digitizing results, we obtained an F1 score of 0.74 (precision is 0.59 and recall is 1.0), showing that the proposed method can effectively map small and irregular thermokarst landforms. It is potentially viable to apply the designed method to mapping diverse thermokarst landforms in a larger area where high-resolution images and training data are available.

Automatic recognition of holistic functional brain networks using iteratively optimized convolutional neural networks (IO-CNN) with weak label initialization

fMRI data decomposition techniques have advanced significantly from shallow models such as Independent Component Analysis (ICA) and Sparse Coding and Dictionary Learning (SCDL) to deep learning models such Deep Belief Networks (DBN) and Convolutional Autoencoder (DCAE). However, interpretations of those decomposed networks are still open questions due to the lack of functional brain atlases, no correspondence across decomposed or reconstructed networks across different subjects, and significant individual variabilities. Recent studies showed that deep learning, especially deep convolutional neural networks (CNN), has extraordinary ability of accommodating spatial object patterns, e.g., our recent works using 3D CNN for fMRI-derived network classifications achieved high accuracy with a remarkable tolerance for mistakenly labelled training brain networks. However, the training data preparation is one of the biggest obstacles in these supervised deep learning models for functional brain network map recognitions, since manual labelling requires tedious and time-consuming labours which will sometimes even introduce label mistakes. Especially for mapping functional networks in large scale datasets such as hundreds of thousands of brain networks used in this paper, the manual labelling method will become almost infeasible. In response, in this work, we tackled both the network recognition and training data labelling tasks by proposing a new iteratively optimized deep learning CNN (IO-CNN) framework with an automatic weak label initialization, which enables the functional brain networks recognition task to a fully automatic large-scale classification procedure. Our extensive experiments based on ABIDE-II 1099 brains' fMRI data showed the great promise of our IO-CNN framework.

Performance prediction and adaptation for database management system workload using Case-Based Reasoning approach

Workload management in a Database Management System (DBMS) has become difficult and challenging because of workload complexity and heterogeneity. During and after execution of the workload, it is hard to control and handle the workload. Before executing the workload, predicting its performance can help us in workload management. By knowing the type of workload in advance, we can predict its performance in an adaptive way that will enable us to monitor and control the workload, which ultimately leads to performance tuning of the DBMS. This study proposes a predictive and adaptive framework named as the Autonomic Workload Performance Prediction (AWPP) framework. The proposed AWPP framework predicts and adapts the DBMS workload performance on the basis of information available in advance before executing the workload. The Case-Based Reasoning (CBR) approach is used to solve the workload management problem. The proposed CBR approach is compared with other machine learning techniques. To validate the AWPP framework, a number of benchmark workloads of the Decision Support System (DSS) and the Online Transaction Processing (OLTP) are executed on the MySQL DBMS. For preparation of training and testing data, we executed more than 1000 TPC-H and TPC-C like workloads on a standard data set. The results show that our proposed AWPP framework through CBR modeling performs better in predicting and adapting the DBMS workload. DBMSs algorithms can be optimized for this prediction and workload can be controlled and managed in a better way. In the end, the results are validated by performing post-hoc tests.

Neural network based automated detection of link failures in wireless sensor networks and extension to a study on the detection of disjoint nodes

In the broad research area of wireless sensor networks (WSN), detection of link failure is still in its infancy. In this paper, we propose to use a neural network model for detection of link failure in WSN. The neural network has been allowed to learn and adapt with the help of gradient descent based learning algorithm. We demonstrate the proposed model with regard to the preparation of training data and implementation of the model. This paper also provides a thorough theoretical and analytical investigation of link failures in WSN. The proposed neural network based model has been evaluated carefully with regard to testbed experiments. The simulation-based experiment has been conducted to justify the applicability of the proposed model for dense networks that could contain around 1000 links. We also analyze the theoretical performance of the proposed neural network based algorithm with regard to various performance evaluative measures such as failure detection accuracy, false alarm rate. The simulated experiments, as well as the testbed experiments in indoor and outdoor environments, suggest that the method is capable of link failure detection with higher detection rate and it is consistent. Furthermore, this article also reports a comprehensive case study as an extension of this present research towards automated detection of disjoint and disconnected nodes in a sensor network.

Voltage Stability Prediction Using Active Machine Learning

An active machine learning technique for monitoring the voltage stability in transmission systems is presented. It has been shown that machine learning algorithms may be used to supplement the traditional simulation approach, but they suffer from the difficulties of online machine learning model update and offline training data preparation. We propose an active learning solution to enhance existing machine learning applications by actively interacting with the online prediction and offline training process. The technique identifies operating points where machine learning predictions based on power system measurements contradict with actual system conditions. By creating the training set around the identified operating points, it is possible to improve the capability of machine learning tools to predict future power system states. The technique also accelerates the offline training process by reducing the amount of simulations on a detailed power system model around operating points where correct predictions are made. Experiments show a significant advantage in relation to the training time, prediction time, and number of measurements that need to be queried to achieve high prediction accuracy.

Text Classification for Organizational Researchers: A Tutorial.

Organizations are increasingly interested in classifying texts or parts thereof into categories, as this enables more effective use of their information. Manual procedures for text classification work well for up to a few hundred documents. However, when the number of documents is larger, manual procedures become laborious, time-consuming, and potentially unreliable. Techniques from text mining facilitate the automatic assignment of text strings to categories, making classification expedient, fast, and reliable, which creates potential for its application in organizational research. The purpose of this article is to familiarize organizational researchers with text mining techniques from machine learning and statistics. We describe the text classification process in several roughly sequential steps, namely training data preparation, preprocessing, transformation, application of classification techniques, and validation, and provide concrete recommendations at each step. To help researchers develop their own text classifiers, the R code associated with each step is presented in a tutorial. The tutorial draws from our own work on job vacancy mining. We end the article by discussing how researchers can validate a text classification model and the associated output.

DEVELOPMENT OF TIME-SERIES HUMAN SETTLEMENT MAPPING SYSTEM USING HISTORICAL LANDSAT ARCHIVE

Methodology of automated human settlement mapping is highly needed for utilization of historical satellite data archives for urgent issues of urban growth in global scale, such as disaster risk management, public health, food security, and urban management. As development of global data with spatial resolution of 10-100 m was achieved by some initiatives using ASTER, Landsat, and TerraSAR-X, next goal has targeted to development of time-series data which can contribute to studies urban development with background context of socioeconomy, disaster risk management, public health, transport and other development issues. We developed an automated algorithm to detect human settlement by classification of built-up and non-built-up in time-series Landsat images. A machine learning algorithm, Local and Global Consistency (LLGC), was applied with improvements for remote sensing data. The algorithm enables to use MCD12Q1, a MODIS-based global land cover map with 500-m resolution, as training data so that any manual process is not required for preparation of training data. In addition, we designed the method to composite multiple results of LLGC into a single output to reduce uncertainty. The LLGC results has a confidence value ranging 0.0 to 1.0 representing probability of built-up and non-built-up. The median value of the confidence for a certain period around a target time was expected to be a robust output of confidence to identify built-up or non-built-up areas against uncertainties in satellite data quality, such as cloud and haze contamination. Four scenes of Landsat data for each target years, 1990, 2000, 2005, and 2010, were chosen among the Landsat archive data with cloud contamination less than 20%.We developed a system with the algorithms on the Data Integration and Analysis System (DIAS) in the University of Tokyo and processed 5200 scenes of Landsat data for cities with more than one million people worldwide.

DEVELOPMENT OF TIME-SERIES HUMAN SETTLEMENT MAPPING SYSTEM USING HISTORICAL LANDSAT ARCHIVE

Abstract. Methodology of automated human settlement mapping is highly needed for utilization of historical satellite data archives for urgent issues of urban growth in global scale, such as disaster risk management, public health, food security, and urban management. As development of global data with spatial resolution of 10-100 m was achieved by some initiatives using ASTER, Landsat, and TerraSAR-X, next goal has targeted to development of time-series data which can contribute to studies urban development with background context of socioeconomy, disaster risk management, public health, transport and other development issues. We developed an automated algorithm to detect human settlement by classification of built-up and non-built-up in time-series Landsat images. A machine learning algorithm, Local and Global Consistency (LLGC), was applied with improvements for remote sensing data. The algorithm enables to use MCD12Q1, a MODIS-based global land cover map with 500-m resolution, as training data so that any manual process is not required for preparation of training data. In addition, we designed the method to composite multiple results of LLGC into a single output to reduce uncertainty. The LLGC results has a confidence value ranging 0.0 to 1.0 representing probability of built-up and non-built-up. The median value of the confidence for a certain period around a target time was expected to be a robust output of confidence to identify built-up or non-built-up areas against uncertainties in satellite data quality, such as cloud and haze contamination. Four scenes of Landsat data for each target years, 1990, 2000, 2005, and 2010, were chosen among the Landsat archive data with cloud contamination less than 20%.We developed a system with the algorithms on the Data Integration and Analysis System (DIAS) in the University of Tokyo and processed 5200 scenes of Landsat data for cities with more than one million people worldwide.

A semantic annotation framework for scientific publications

Considering the growing volume of scientific literature, techniques that enable automatic detection of informational entities existing in scientific research articles may contribute to the extension of scientific knowledge and practical usages. Although there have been several efforts to extract informative entities from patent and biomedical research articles, there are few attempts in other scientific literatures. In this paper, we introduce an automatic semantic annotation framework for research articles based on entity recognition techniques. Our approach includes tag set modeling for semantic annotation, semi-automatic annotation tool, manual annotation for training data preparation, and supervised machine learning to develop entity type recognition module. For experiments, we choose two different domains, such as information and communication technology and chemical engineering due to their high usages. In addition, we provide three application scenarios of how our annotation framework can be used and extended further. It is to guide potential researchers who are willing to link their own contents with external data.