Clean Dataset Research Articles

Enhanced Unsupervised Image Generation using GAN based Convolutional Nets

Generative Adversarial Networks (GANs) use deep learning methods like neural nets for generative modeling. Neural style transferring of images and facial character generation of anime images are previously implemented by applying GAN methods but were not successful in giving a promising output. In this work, Image Processing is applied on the datasets in the mode along with the training of GAN system. The problem of applying GAN to generate specific images is addressed by using a clean and problem specific dataset for anime facial character generation. Modeling is done by applying Convolutional Neural Nets, GANs empirically. Neural style transfer, Automatic Anime characters are generated with high-resolution, and this model tackles the limitations by progressively increasing the resolution of both generated images and structural conditions during training. This model can be used to develop unique anime characters or the image generated can be used as inspiration by artists and graphic designers, can be used as filters in famous apps such as snapchat for style transferring. With different evaluations and result analysis, it is observed that this model is a stable and high-quality model.

A methodology for conducting efficient sanitization of HTTP training datasets

The performance of anomaly-based intrusion detection systems depends on the quality of the datasets used to form normal activity profiles. Suitable datasets should include high volumes of real-life data free from attack instances. On account of this requirement, obtaining quality datasets from collected data requires a process of data sanitization that may be prohibitive if done manually, or uncertain if fully automated.In this work, we propose a sanitization approach for obtaining datasets from HTTP traces suited for training, testing, or validating anomaly-based attack detectors. Our methodology has two sequential phases. In the first phase, we clean known attacks from data using a pattern-based approach that relies on tools that detect URI-based known attacks. In the second phase, we complement the result of the first phase by conducting assisted manual labeling systematically and efficiently, setting the focus of expert examination not on the raw data (which would be millions of URIs), but on the set of words that compose the URIs. This dramatically downsizes the volume of data that requires expert discernment, making manual sanitization of large datasets feasible. We have applied our method to sanitize a trace that includes 45 million requests received by the library web server of the University of Seville. We were able to generate clean datasets in less than 84 h with only 33 h of manual supervision. We have also applied our method to some public benchmark datasets, confirming that attacks unnoticed by signature-based detectors can be discovered in a reduced time span.

A Cautionary Tale for Machine Learning Design: why we Still Need Human-Assisted Big Data Analysis

Supervised Machine Learning (ML) requires that smart algorithms scrutinize a very large number of labeled samples before they can make right predictions. And this is not always true either. In our experience, in fact, a neural network trained with a huge database comprised of over fifteen million water meter readings had essentially failed to predict when a meter would malfunction/need disassembly based on a history of water consumption measurements. With a second step, we developed a methodology, based on the enforcement of a specialized data semantics, that allowed us to extract only those samples for training that were not noised by data impurities. With this methodology, we re-trained the neural network up to a prediction accuracy of over 80%. Yet, we simultaneously realized that the new training dataset was significantly different from the initial one in statistical terms, and much smaller, as well. We had reached a sort of paradox: We had alleviated the initial problem with a better interpretable model, but we had changed the replicated form of the initial data. To reconcile that paradox, we further enhanced our data semantics with the contribution of field experts. This has finally led to the extrapolation of a training dataset truly representative of regular/defective water meters and able to describe the underlying statistical phenomenon, while still providing an excellent prediction accuracy of the resulting classifier. At the end of this path, the lesson we have learnt is that a human-in-the-loop approach may significantly help to clean and re-organize noised datasets for an empowered ML design experience.

Stability analysis of feature ranking techniques in the presence of noise: a comparative study

Noisy data is one of the common problems associated with real-world data, and may affects the performance of the data models, consequent decisions and the performance of feature ranking techniques. In this paper, we show how stability performance can be changed if different feature ranking methods against attribute noise and class noise are used. We consider Kendall's Tau rank correlation and Spearman rank correlation to evaluate various feature ranking methods stability, and quantify the degree of agreement between ordered lists of features created by a filter on a clean dataset and its outputs on the same dataset corrupted with different combinations of the noise level. According to the results of Kendall and Spearman measures, Gini index (GI) and information gain (IG) have the best performances respectively. Nevertheless, both Kendall and Spearman measures results show that ReliefF (RF) is the most sensitive (the worst) performance.

Distributed Approximate Newton's Method Robust to Byzantine Attackers

There is a recent surge of interest in the design of the first-order and the second-order distributed machine learning algorithms. However, distributed algorithms are sensitive to Byzantine attackers who can send falsified information to prevent the convergence of algorithms or lead the algorithms to converge to value of the attackers' choice. Some recent works have proposed algorithms that can defend against Byzantine attackers for the first-order methods. In this paper, we design two algorithms that can deal with Byzantine attackers for the second-order methods. The main idea of the first algorithm, named median-based approximate Newton's method (MNM), is to ask the parameter server to aggregate gradient information and approximate Newton's direction from all workers by geometric median. We show that MNM can converge when up to half of the workers are Byzantine attackers. To deal with the case with an arbitrary number of attackers, we then propose a comparison-based approximate Newton's method (CNM). The main idea of CNM is to ask the server to randomly select a small clean dataset and compute noisy gradient and Newton's direction using this small dataset. These noisy information will then be used as an approximation of the ground truth to filter out bad information from Byzantine attackers. We show that CNM can converge to the neighborhood of the population minimizer even when more than half of the workers are Byzantine workers. We further provide numerical examples to illustrate the performance of the proposed algorithms.

Muon Hunter: a Zooniverse project

The large datasets and often low signal-to-noise inherent to the raw data of modern astroparticle experiments calls out for increasingly sophisticated event classification techniques. Machine learning algorithms, such as neural networks, have the potential to outperform traditional analysis methods, but come with the major challenge of identifying reliably classified training samples from real data. Citizen science represents an effective approach to sort through the large datasets efficiently and meet this challenge. Muon Hunter is a project hosted on the Zooniverse platform, wherein volunteers sort through pictures of data from the VERITAS cameras to identify muon ring images. Each image is classified multiple times to produce a clean dataset used to train and validate a convolutional neural network model both able to reject background events and identify suitable calibration data to monitor the telescope performance as a function of time.

Learning From Large-Scale Noisy Web Data With Ubiquitous Reweighting for Image Classification.

Many important advances of deep learning techniques have originated from the efforts of addressing the image classification task on large-scale datasets. However, the construction of clean datasets is costly and time-consuming since the Internet is overwhelmed by noisy images with inadequate and inaccurate tags. In this paper, we propose a Ubiquitous Reweighting Network (URNet) that can learn an image classification model from noisy web data. By observing the web data, we find that there are five key challenges, i.e., imbalanced class sizes, high intra-classes diversity and inter-class similarity, imprecise instances, insufficient representative instances, and ambiguous class labels. With these challenges in mind, we assume every training instance has the potential to contribute positively by alleviating the data bias and noise via reweighting the influence of each instance according to different class sizes, large instance clusters, its confidence, small instance bags, and the labels. In this manner, the influence of bias and noise in the data can be gradually alleviated, leading to the steadily improving performance of URNet. Experimental results in the WebVision 2018 challenge with 16 million noisy training images from 5000 classes show that our approach outperforms state-of-the-art models and ranks first place in the image classification task.

Deep Learning Predictive Models for Cognitive Radio System

Cognitive Radio (CR) was introduced to improve the utilization of Radio Frequencies (RF) that remain under-utilized by the primary users (licensee). The main idea behind CR is to allow un-licensed (secondary) users to occupy vacancies in licensed bands. However, CR mandates the secondary user to vacate the frequency band within a specified time after the primary user attempts to use the frequency band. CR does not expect the primary users to share their frequency usage schedules and hence the secondary users have to scan and predict the vacancy. The advantage for the secondary users is that they do not pay for utilization of band, if they are conformal to the CR specifications. CR is the next generation of smart communication systems. CR requires continuous monitoring of the intended RF band in the intended geographical area. This information may be used to predict spectral vacancies (white spaces). Certain bands, e.g. Analog TV bands, will have pre declared utilization schedules but in general, spectrum utilization is a random process and hence prediction can be difficult. However, Deep Learning (DL) techniques can improve the accuracy of prediction. Deep Learning techniques require large and clean data sets to work correctly. Such data sets are also necessary to compare achievable accuracy of prediction algorithms. Towards this end, we have created data sets that can be used for simulation, training and testing of CR over GSM band (890-960MHz). A typical file with two hour of observations will have about 1.2 million samples. More than 1000 sets of data samples have been captured from urban and rural areas in India. All the data sets have been cleaned to avoid instrument errors and statistical outliers. In this paper we have used these standardized data sets to perform a comparative analysis of three DL methods for CR, viz. Auto-encoder (AE), Long Short-Term Memory (LSTM) and Multi Layer Perceptron (MLP). Results of the comparison are discussed.

Recognition of Degraded and Non Degraded Roman Characters Using Different Classifiers

In this paper, we present an efficient and robust technique for the recognition of offline roman characters. The main strategy is to extract statistical and similarity features using a combination of grey level co-occurrence matrix (GLCM) and complementary similarity measure (CSM) method. In this work, the CSM method is used to extract features from binary images and combined with GLCM to boost the accuracy of character recognition. The recognition has been done using four different classifiers i.e. artificial neural network (ANN), Naive Bayes classifier, random forest (RF) and support vector machine (SVM). The standard dataset has been used for experimental work. We have done experiments on the clean and noisy dataset. It achieves the accuracy of 100% for some characters without noise and 94:11% with impulsive noise. A comparison of these four classifiers is recorded with and without a noisy environment. On a clean dataset, the random forest provides the best average recognition accuracy of 84:9% for all characters. On low noise datasets, random forest and artificial neural networks have almost the same recognition accuracy and on high noise datasets, SVM provides the highest recognition accuracy.

Distributed Gradient Descent Algorithm Robust to an Arbitrary Number of Byzantine Attackers

Due to the growth of modern dataset size and the desire to harness computing power of multiple machines, there is a recent surge of interest in the design of distributed machine learning algorithms. However, distributed algorithms are sensitive to Byzantine attackers who can send falsified data to prevent the convergence of algorithms or lead the algorithms to converge to value of the attackers' choice. Some recent work proposed interesting algorithms that can deal with the scenario when up to half of the workers are compromised. In this paper, we propose a novel algorithm that can deal with an arbitrary number of Byzantine attackers. The main idea is to ask the parameter server to randomly select a small clean dataset and compute noisy gradient using this small dataset. This noisy gradient will then be used as a ground truth to filter out information sent by compromised workers. We show that the proposed algorithm converges to the neighborhood of the population minimizer regardless the number of Byzantine attackers. We further provide numerical examples to show that the proposed algorithm can benefit from the presence of good workers and achieve better performance than existing algorithms.

Modeling the thermal and soot oxidation dynamics inside a ceria-coated gasoline particulate filter

Gasoline particulate filters (GPFs) are practically adoptable devices to mitigate particulate matter emissions from vehicles using gasoline direct ignition engines. This paper presents a newly developed control-oriented model to characterize the thermal and soot oxidation dynamics in a ceria-coated GPF. The model utilizes the GPF inlet exhaust gas temperature, exhaust gas mass flow rate, the initial GPF soot loading density, and air–fuel ratio to predict the internal GPF temperature and the amount of soot oxidized during regeneration events. The reaction kinetics incorporated in the model involve the rates of both oxygen- and ceria-initiated soot oxidation reactions. Volumetric model parameters are calculated from the geometric information of the coated GPF, while the air–fuel ratio is used to determine the volume fractions of the exhaust gas constituents. The exhaust gas properties are evaluated using the volume fractions and thermodynamic tables, while the cordierite specific heat capacity is identified using a clean experimental data set. The enthalpies of the regeneration reactions are calculated using thermochemical tables. Physical insights from the proposed model are thus enhanced by limiting the number of parameters obtained from fitting to only those which cannot be directly measured from experiments. The parameters of the model are identified using the particle swarm optimization algorithm and a cost function designed to simultaneously predict both thermal and soot oxidation dynamics. Parameter identification and model validation are performed using independent data sets from laboratory experiments conducted on a ceria-coated GPF. This work demonstrates that the proposed model can be successfully implemented to predict ceria-coated GPF dynamics under different soot loading and temperature conditions.

Natural Neighbor Algorithm for Breast Cancer Diagnosis

Machine learning to diagnose breast cancer is an important, real-world medical problem. As one of the most popular machine learning algorithm, K-Nearest Neighbors (KNN) algorithm is widely used in breast cancer classification and can provide high classification accuracy and effective diagnostic capabilities. However, selection of parameter K is still an unsolved problem for K-Nearest Neighbors (KNN). To address the problem, we introduce a novel neighbor form, Natural Neighbor (NaN), which is obtained adaptively by its search algorithm. We firstly use a noise filter, which called edited natural neighbor algorithm (ENaN), to eliminate the noises and global outliers of Wisconsin’s breast cancer prognosis dataset. Then, cleaned data set and natural neighbor algorithm are used to construct a new non parameter diagnostic system. The main advantages of the proposed algorithm are that it does not need any parameters and can maintain high classification accuracy. Experiments show that the classification accuracy of proposed method is similar to the highest classification accuracy obtained by traditional K-Nearest Neighbors algorithm with different K values.

Linear Correlations between Adsorption Energies and HOMO Levels for the Adsorption of Small Molecules on TiO2 Surfaces

Adsorption is a fundamental step in catalysis on a solid surface, and therefore, its understanding is important for explaining its behavior. This work investigated the adsorption of various small molecules, including H2, N2, CO, CO2, CH4, NH3, H2O, H2S, dimethyl sulfoxide, alkanes, alkenes, alkynes, aromatic compounds, alcohols, aldehydes, ketones, nitriles, carboxylic acids, amides, and amines, on the anatase (101) and rutile (110) surfaces of TiO2, using periodic density functional theory calculations and statistical methods. Adsorption energies were computed at the same level of theory to obtain a clean and consistent data set. A linear relationship was observed between the adsorption energies of these molecules and their highest occupied molecular orbital (HOMO) levels, whereas no obvious correlation was evident for the lowest unoccupied molecular orbital levels. Improved correlations between the adsorption energies and HOMO levels were generated by dividing these molecules into two subgroups: hydrocarbons and heteroatom-containing compounds. Interactions between frontier molecular orbitals and the surfaces were considered, to gain a better understanding of the significant correlations that were identified. The data show that these relationships can be primarily ascribed to the interactions between the HOMO of the small molecule and conduction state of the TiO2 surface. The statistical analysis using machine learning demonstrated that the HOMO and dipole moment are the first and second most important properties, respectively, in terms of rationalizing and predicting the adsorption energies.

Facilitating reproducible project management and manuscript development in team science: The projects R package.

The contemporary scientific community places a growing emphasis on the reproducibility of research. The projects R package is a free, open-source package created in the interest of facilitating reproducible research workflows. It adds to existing software tools for reproducible research and introduces several practical features that are helpful for scientists and their collaborative research teams. For each individual project, it supplies a framework for storing raw and cleaned study data sets, and it provides script templates for protocol creation, data cleaning, data analysis and manuscript development. Internal databases of project and author information are generated and displayed, and manuscript title pages containing author lists and their affiliations are automatically generated from the internal database. File management tools allow teams to organize multiple projects. When used on a shared file system, multiple researchers can harmoniously contribute to the same project in a less punctuated manner, reducing the frequency of misunderstandings and the need for status updates.

A Training Data Set Cleaning Method by Classification Ability Ranking for the k -Nearest Neighbor Classifier.

The k -nearest neighbor (KNN) rule is a successful technique in pattern classification due to its simplicity and effectiveness. As a supervised classifier, KNN classification performance usually suffers from low-quality samples in the training data set. Thus, training data set cleaning (TDC) methods are needed for enhancing the classification accuracy by cleaning out noisy, or even wrong, samples in the original training data set. In this paper, we propose a classification ability ranking (CAR)-based TDC method to improve the performance of a KNN classifier, namely CAR-based TDC method. The proposed classification ability function ranks a training sample in terms of its contribution to correctly classify other training samples as a KNN through the leave-one-out (LV1) strategy in the cleaning stage. The training sample that likely misclassifies the other samples during the KNN classifications according to the LV1 strategy is considered to have lower classification ability and will be cleaned out from the original training data set. Extensive experiments, based on ten real-world data sets, show that the proposed CAR-based TDC method can significantly reduce the classification error rates of KNN-based classifiers, while reducing computational complexity thanks to a smaller cleaned training data set.

External prior learning and internal mean sparse coding for image denoising

Image prior and sparse coding learning methods have important uses in image denoising. Many denoising methods learn priors either from the noisy image itself or an external clean image dataset. But using only these as priors does not always reconstruct the image effectively. In addition, when the image is corrupted by noise, the local sparse coding coefficient obtained from a noisy image patch is inaccurate, restricting denoising performance. We present a noise removal framework based on external prior learning and an internal mean sparse coding method, making use of the innate sparsity and nonlocal self-similarity (NSS) of natural images. Specifically, we first obtain external priors from a clean natural image dataset by Gaussian mixture model. The external priors are applied to guide the subspace clustering of internal noisy image patches, and a compact dictionary is generated for each internal noisy patch cluster. Then an internal mean sparse coding strategy based on NSS is introduced into the sparse representation model, whose regularization parameters then are deduced through a Bayesian framework. An iterative shrinkage method is employed to solve the l1-optimization problem in the sparse representation model. Application of the noise removal model to 16 test images demonstrates denoising performance exceeding other competing methods.

Unsupervised multi-view non-negative for law data feature learning with dual graph-regularization in smart Internet of Things

Abstract In the real world, the law data in the smart Internet of Things usually consists of heterogeneous information with some noises. Non-negative matrix factorization is a popular tool for multi-view learning, which can be employed to represent and learn heterogeneous law features comprehensively. However, current NMF-based techniques generally use clean multi-view datasets to generate common subspace, while in practice, they often contain some noises or unrelated items so that the performance of the algorithms may be severely degraded. In this paper, we propose to develop a novel subspace learning model, called Adaptive Dual Graph-regularized Multi-View Non-Negative Feature Learning (ADMFL), for multi-view data representation. We utilize the geometric structures of both data and feature manifold to model the distribution of data points in the common subspace. Meanwhile, we lift the effect of unrelated features down through separating the view-specific features for each view. Moreover, we introduce a weight factor for all views and maintain the sparsity of the latent common representation. An effective objective function is thus designed and iteratively updated until convergence. Experiments on standard datasets demonstrate that the proposed ADMFL method outperforms other compared methods in the paper.

To clean or not to clean phenotypic datasets for outlier plants in genetic analyses?

Based on case studies, we discuss the extent to which genome-wide association studies (GWAS) are affected by outlier plants, i.e. those deviating from the expected distribution on a multi-criteria basis. Using a raw dataset consisting of daily measurements of leaf area, biomass, and plant height for thousands of plants, we tested three different cleaning methods for their effects on genetic analyses. No-cleaning resulted in the highest number of dubious quantitative trait loci, especially at loci with highly unbalanced allelic frequencies. A trade-off was identified between the risk of false-positives (with no-cleaning and/or a low threshold for minor allele frequency) and the risk of missing interesting rare alleles. Cleaning can lower the risk of the latter by making it possible to choose a higher threshold in GWAS.

DECODE: Deep Confidence Network for Robust Image Classification.

Recent years have witnessed the success of deep convolutional neural networks for image classification and many related tasks. It should be pointed out that the existing training strategies assume that there is a clean dataset for model learning. In elaborately constructed benchmark datasets, deep network has yielded promising performance under the assumption. However, in real-world applications, it is burdensome and expensive to collect sufficient clean training samples. On the other hand, collecting noisy labeled samples is very economical and practical, especially with the rapidly increasing amount of visual data in the web. Unfortunately, the accuracy of current deep models may drop dramatically even with 5%-10% label noise. Therefore, enabling label noise resistant classification has become a crucial issue in the data driven deep learning approaches. In this paper, we propose a DEep COnfiDEnce network (DECODE) to address this issue. In particular, based on the distribution of mislabeled data, we adopt a confidence evaluation module that is able to determine the confidence that a sample is mislabeled. With the confidence, we further use a weighting strategy to assign different weights to different samples so that the model pays less attention to low confidence data, which is more likely to be noise. In this way, the deep model is more robust to label noise. DECODE is designed to be general, such that it can be easily combined with existing studies. We conduct extensive experiments on several datasets, and the results validate that DECODE can improve the accuracy of deep models trained with noisy data.

Stability analysis of Feature Ranking Techniques in the presence of noise: A comparative study

Noisy data is one of the common problems associated with real-world data, and may affects the performance of the data models, consequent decisions and the performance of feature ranking techniques. In this paper, we show how stability performance can be changed if different feature ranking methods against attribute noise and class noise are used. We consider Kendall's Tau rank correlation and Spearman rank correlation to evaluate various feature ranking methods stability, and quantify the degree of agreement between ordered lists of features created by a filter on a clean dataset and its outputs on the same dataset corrupted with different combinations of the noise level. According to the results of Kendall and Spearman measures, Gini index (GI) and information gain (IG) have the best performances respectively. Nevertheless, both Kendall and Spearman measures results show that ReliefF (RF) is the most sensitive (the worst) performance.