Less Amount Of Data Research Articles

A fast and time-efficient machine learning approach to dark matter searches in compressed mass scenario

Various analyses for searching for the signature of SUSY or exotic particles have been carried out by the experiments at CERN. These analyses made use of traditional cut and count methods. While this method has yielded promising results, it has been challenging in the region where the mass difference between SUSY particles is small. Deep learning is currently widely employed in most data analysis tasks, including high energy physics, and has made significant advances in almost all fields for collecting and interpreting huge data samples. In this paper, a fast and time-efficient classification technique is proposed, utilizing machine learning algorithms to distinguish dark matter signal from SM background in compressed mass spectra scenarios at a center-of-mass energy of 14 TeV. A classification model was built in a short amount of time using 2D histograms produced with less amount of data, effectively reducing computational costs through the transfer learning of pre-trained deep models while maintaining a high level of classification accuracy.

Lightning Nowcasting Using Solely Lightning Data

Lightning is directly or indirectly responsible for significant human casualties and property damage worldwide. A timely prediction of its occurrence can enable authorities and the public to take necessary precautionary actions resulting in diminishing the potential hazards caused by lightning. In this paper, based on the assumption that atmospheric phenomena behave in a continuous manner, we present a model based on residual U-nets where the network architecture leverages this inductive bias by combining information passing directly from the input to the output with the necessary required changes to the former, predicted by a neural network. Our model is trained solely on lightning data from geostationary weather satellites and can be used to predict the occurrence of future lightning. Our model has the advantage of not relying on numerical weather models, which are inherently slow due to their sequential nature, enabling it to be used for near-future prediction (nowcasting). Moreover, our model has similar performance compared to other machine learning based lightning predictors in the literature while using significantly less amount of data for training, limited to lightning data. Our model, which is trained for four different lead times of 15, 30, 45, and 60 min, outperforms the traditional persistence baseline by 4%, 12%, and 22% for lead times of 30, 45, and 60 min, respectively, and has comparable accuracy for 15 min lead time.

Driver Vigilance Detection Based on Limited EEG Signals

Driver vigilance detection is an important topic for traffic safety improvement and brain-computer interface design. Electroencephalography (EEG) can directly provide real-time information and is sensitive to the change of vigilance. Thus, EEG signals are widely used for driver vigilance detection. Previous studies commonly acquired EEG signals from a large number of electrode channels and utilized all the collected data for recognition tasks, which may introduce a large amount of noise and redundant data and limit their application potentials in practical systems. This paper examines the potential of using EEG signals from only one frequency band or from only a small subset of related electrode channels in recognizing driver vigilance state. The widely used SEED VIG dataset is adopt for experiments. Four candidate sets of EEG frequency bands and four candidate sets of EEG electrode channels are determined according to the statistical significance results of EEG power spectral density (PSD) features and differential entropy (DE) features. The experimental results show that the recognition accuracy is higher when using EEG signals from the selected frequency band (i.e., alpha band) or the selected electrodes (i.e., T7, TP7, CP1) than when using all the data. Specifically, the best recognition accuracy is 91.31% for the alpha band and 76.81% for the selected electrodes. These results indicate that higher driver vigilance recognition accuracy can be achieved with much less amount of data, which would facilitate the development of wearable equipment based on EEG signals.

Incidence, Risk Factors, and Prognosis of Hidradenitis Suppurativa Across the Globe: Insights from the Literature.

Hidradenitis suppurativa, a chronic inflammatory disease of the skin, affects a patient's quality of life to a greater extent. While the disease burden, including its incidence and prevalence, has been extensively studied in the western population, there is a paucity of data from developing countries on the epidemiology of Hidradenitis suppurativa. Therefore, a general literature review was conducted to shed light on the epidemiology of Hidradenitis suppurativa across the globe. We reviewed the most recently available information on epidemiology, including incidence, prevalence, risk factors, prognosis and quality of life, complications, and associated comorbid among patients with Hidradenitis suppurativa. The estimated global prevalence of Hidradenitis suppurativa is found to be 0.00033-4.1%, with a relatively higher prevalence of 0.7-1.2% in European and US populations. Both genetic and environmental factors are associated with Hidradenitis suppurativa. Patients with Hidradenitis suppurativa have associated comorbid such as cardiovascular disease, type II diabetes mellitus, mental health issues, and sleep and sexual dysfunctions. These patients spend poor quality of life and tend to be less productive. Future studies are needed to assess the burden of Hidradenitis suppurativa in developing countries. Since the disease tends to be underdiagnosed, future studies should rely on clinical diagnosis rather than self-reporting to avoid the potential of recall bias. Attention needs to be diverted to developing countries with less amount of data on Hidradenitis suppurativa.

3D Cardiac Substructures Segmentation from CMRI using Generative Adversarial Network (GAN).

Cardiac magnetic resonance imaging (CMRI) improves the diagnosis of cardiovascular diseases by providing images at high spatio-temporal resolution helping physicians in providing correct treatment plans. Segmentation and identification of various substructures of the heart at different cardiac phases of end-systole and end-diastole helps in the extraction of ventricular function information such as stroke volume, ejection fraction, myocardium thickness, etc. Manual delineation of the substructures is tedious, time-consuming, and error-prone. We have implemented a 3D GAN that includes 3D contextual information capable of segmenting and identifying the substructures at different cardiac phases with improved accuracy. Our method is evaluated on the ACDC dataset (4 pathologies, 1 healthy group) to show that the proposed out-performs other methods in literature with less amount of data. Also, the proposed provided a better Dice score in segmentation surpassing other methods on a blind-tested M&Ms dataset.

Using Movie Genres in Neural Network Based Collaborative Filtering Movie Recommendation System to Reduce Cold Start Problem

In a recommendation system, a cold start problem occurs when the system is unable to make any inferences for users or items for which it has not yet acquired adequate data. This research proposes a modified version of the Neural Collaborative Filtering model that has revolutionized collaborative filtering recommendation system with its implementation of neural network in the model. The proposed model is called Neural Collaborative Filtering Modified (NCFM). NCFM addresses the cold start problem by reducing the model dependency on users and items interactions with the integration of item feature (movie genres in this case) before the concatenation step in the NCF model. Hit ratio@10 and normalized discounted cumulative gain@10 were used as the evaluation method because they closely represents real-world recommendation scenario. Performance of both models were tested and compared using two movie rating dataset MovieLens1M provided by GroupLens and a modified version of MovieLens1M with significantly less amount of data to simulate the cold start problem called MovieLensCold. The results shows that NCFM can outperforms NCF by more than two times over or around 133% performance increase on MovieLensCold and around 15% increase on MovieLens1M dataset. Keywords— movie recommendation system, cold start problem, collaborative filtering, NCFM, neural network based recommendation system

FIDOOP: PARALLEL MINING OF FREQUENT ITEM SETS USING MAPREDUCE

Due to the exponential increase of real-time data monitoring systems, the extraction of frequent item (Frequent item set mining) set from large uncertain database is the challenging task. The existing parallel mining algorithm for frequent item sets includes the limitations in terms of more memory usage and excessive run time even for less amount of data. To overcome this problem, the FiDoop based item set mining algorithm is proposed by using map reduce framework. It is used to improve the performance of load balancing operation in an uncertain database for computing frequent patterns. This system includes data uploading, preprocessing, threshold, find support and confidence, merge and result. Initially, the data is selected from the dataset and uploaded in the server. Afterwards, the preprocessing stage removes columns which contains unwanted entries. The information is analyzed and partitioned to compute threshold value. These data are classified depends on threshold values and the clustering algorithm is used to find high support and confidence values among clusters to discover frequent item. Finally, those frequent item sets are merged to acquire a frequent pattern. The proposed system is mainly developed for improving the accuracy and it is evaluated based on the performance measures of accuracy, memory usage and execution time.

Square Error Method for threshold estimation in extreme value analysis of wave heights

One of the methods used to model the probability distribution of extreme values is the peaks over threshold method. It requires an appropriate estimation of the threshold above which exceedances are considered to be a sample from a generalized Pareto distribution. Several methods have been proposed to determine the optimal threshold. In this paper, we assess the applicability of four methods developed for hydrological, coastal engineering, financial and stock market purposes, to the case of wave heights from extreme sea conditions due to storms, where significantly less amount of data may be available. Based on the results of this assessment, an improved square error method is formulated, which further allows evaluating the uncertainty in threshold estimation, as well as the uncertainty in the estimation of wave heights for given return periods. The improved method is applied using significant wave height data from hindcast of tropical and extra-tropical storms in the Gulf of Mexico. Results illustrate the capabilities of the proposed method.

Efficient Prediction of Liver Disease using Selected Attributes

Liver plays a vital role in the human body that performs several crucial life functions. A number of liver diseases exist and it is a challenging task to diagnose the liver disease at its early stage. In recent years, several data mining techniques have been used in medical field for prediction but there can be further improvements for quick and accurate diagnose of liver disease. In this paper, a variety of Classifiers have been experimented on Indian liver disease patients dataset which is publicly available on Kaggle. Attribute subset selection is performed to identify significant attributes and the resulting dataset is named as Selected Attributes Dataset (SAD). SAD provides more accuracy in less computation time using Random forest classification algorithm and improved system including these parameters i.e., the efficiency of the system can be increased, early decision making, less time and space required. This research work will provide help to predict liver disease with less amount of data, i.e., number of attributes.

Positive–negative corresponding normalized ghost imaging based on an adaptive threshold

Ghost imaging (GI) technology has attracted increasing attention as a new imaging technique in recent years. However, the signal-to-noise ratio (SNR) of GI with pseudo-thermal light needs to be improved before it meets engineering application demands. We therefore propose a new scheme called positive–negative correspondence normalized GI based on an adaptive threshold (PCNGI-AT) to achieve a good performance with less amount of data. In this work, we use both the advantages of normalized GI (NGI) and positive–negative correspondence GI (P–NCGI). The correctness and feasibility of the scheme were proved in theory before we designed an adaptive threshold selection method, in which the parameter of object signal selection conditions is replaced by the normalizing value. The simulation and experimental results reveal that the SNR of the proposed scheme is better than that of time-correspondence differential GI (TCDGI), avoiding the calculation of the matrix of correlation and reducing the amount of data used. The method proposed will make GI far more practical in engineering applications.

Extraction of speaker-specific excitation information from linear prediction residual of speech

In this paper, through different experimental studies we demonstrate that the excitation component of speech can be exploited for speaker recognition studies. Linear prediction (LP) residual is used as a representation of excitation information in speech. The speaker-specific information in the excitation of voiced speech is captured using the AutoAssociative Neural Network (AANN) models. The decrease in the error during training and recognizing correct speakers during testing demonstrates that the excitation component of speech contains speaker-specific information and is indeed being captured by the AANN models. The study on the effect of different LP orders demonstrates that for a speech signal sampled at 8 kHz, the LP residual extracted using LP order in the range 8–20 best represents the speaker-specific excitation information. It is also demonstrated that the proposed speaker recognition system using excitation information and AANN models requires significantly less amount of data both during training as well as testing, compared to the speaker recognition system using vocal tract information. Finally the speaker recognition studies on NIST 2002 database demonstrates that even though, the recognition performance from the excitation information alone is poor, when combined with evidence from vocal tract information, there is significant improvement in the performance. This result demonstrates the complementary nature of the excitation component of speech.