Short Data Sets Research Articles

Don't Forget the Quantifiable Relationship between Words: Using Recurrent Neural Network for Short Text Topic Discovery

In our daily life, short texts have been everywhere especially since the emergence of social network. There are countless short texts in online media like twitter, online Q&A sites and so on. Discovering topics is quite valuable in various application domains such as content recommendation and text characterization. Traditional topic models like LDA are widely applied for sorts of tasks, but when it comes to short text scenario, these models may get stuck due to the lack of words. Recently, a popular model named BTM uses word co-occurrence relationship to solve the sparsity problem and is proved effectively. However, both BTM and extended models ignore the inside relationship between words. From our perspectives, more related words should appear in the same topic. Based on this idea, we propose a model named RIBS-TM which makes use of RNN for relationship learning and IDF for filtering high-frequency words. Experiments on two real-world short text datasets show great utility of our model.

A new method for determining the optimal lagged ensemble.

We propose a general methodology for determining the lagged ensemble that minimizes the mean square forecast error. The MSE of a lagged ensemble is shown to depend only on a quantity called the cross‐lead error covariance matrix, which can be estimated from a short hindcast data set and parameterized in terms of analytic functions of time. The resulting parameterization allows the skill of forecasts to be evaluated for an arbitrary ensemble size and initialization frequency. Remarkably, the parameterization also can estimate the MSE of a burst ensemble simply by taking the limit of an infinitely small interval between initialization times. This methodology is applied to forecasts of the Madden Julian Oscillation (MJO) from version 2 of the Climate Forecast System version 2 (CFSv2). For leads greater than a week, little improvement is found in the MJO forecast skill when ensembles larger than 5 days are used or initializations greater than 4 times per day. We find that if the initialization frequency is too infrequent, important structures of the lagged error covariance matrix are lost. Lastly, we demonstrate that the forecast error at leads ≥10 days can be reduced by optimally weighting the lagged ensemble members. The weights are shown to depend only on the cross‐lead error covariance matrix. While the methodology developed here is applied to CFSv2, the technique can be easily adapted to other forecast systems.

Time series analysis of ground-based microwave measurements at K- and V-bands to detect temporal changes in water vapor and temperature profiles

Abstract. Ground-based microwave measurements performed at water vapor and oxygen absorption line frequencies are widely used for remote sensing of tropospheric water vapor density and temperature profiles, respectively. Recent work has shown that Bayesian optimal estimation can be used for improving accuracy of radiometer retrieved water vapor and temperature profiles. This paper focuses on using Bayesian optimal estimation along with time series of independent frequency measurements at K- and V-bands. The measurements are used along with statistically significant but short background data sets to retrieve and sense temporal variations and gradients in water vapor and temperature profiles. To study this capability, the Indian Institute of Tropical Meteorology (IITM) deployed a microwave radiometer at Mahabubnagar, Telangana, during August 2011 as part of the Integrated Ground Campaign during the Cloud Aerosol Interaction and Precipitation Enhancement Experiment (CAIPEEX-IGOC). In this study, temperature profiles for the first time have been estimated using short but statistically significant background information so as to improve the accuracy of the retrieved profiles as well as to be able to detect gradients. Estimated water vapor and temperature profiles are compared with those taken from the reanalysis data updated by the Earth System Research Laboratory, National Oceanic and Atmospheric Administration (NOAA), to determine the range of possible errors. Similarly, root mean square errors are evaluated for a month for water vapor and temperature profiles to estimate the accuracy of the retrievals. It is found that water vapor and temperature profiles can be estimated with an acceptable accuracy by using a background information data set compiled over a period of 1 month.

Self-Taught convolutional neural networks for short text clustering

Short text clustering is a challenging problem due to its sparseness of text representation. Here we propose a flexible Self-Taught Convolutional neural network framework for Short Text Clustering (dubbed STC2), which can flexibly and successfully incorporate more useful semantic features and learn non-biased deep text representation in an unsupervised manner. In our framework, the original raw text features are firstly embedded into compact binary codes by using one existing unsupervised dimensionality reduction method. Then, word embeddings are explored and fed into convolutional neural networks to learn deep feature representations, meanwhile the output units are used to fit the pre-trained binary codes in the training process. Finally, we get the optimal clusters by employing K-means to cluster the learned representations. Extensive experimental results demonstrate that the proposed framework is effective, flexible and outperform several popular clustering methods when tested on three public short text datasets.

Using Word Embedding and Ensemble Learning for Highly Imbalanced Data Sentiment Analysis in Short Arabic Text

Abstract: Sentiment analysis has gained increasing importance with the massive increase of online content. Although several studies have been conducted for western languages, not much has been done for the Arabic language. The purpose of this study is to compare the performance of different classifiers for polarity determination in highly imbalanced short text datasets using features learned by word embedding rather than hand-crafted features. Several base classifiers and ensembles have been investigated with and without SMOTE (Synthetic Minority Over-sampling Technique). Using a dataset of tweets in dialectical Arabic, the results show that applying word embedding with ensemble and SMOTE can achieve more than 15% improvement on average in F 1 score over the baseline, which is a weighted average of precision and recall and is considered a better performance measure than accuracy for imbalanced datasets.

Energy Associated Tuning Method for Short-Term Series Forecasting by Complete and Incomplete Datasets

AbstractThis article presents short-term predictions using neural networks tuned by energy associated to series based-predictor filter for complete and incomplete datasets. A benchmark of high roughness time series from Mackay Glass (MG), Logistic (LOG), Henon (HEN) and some univariate series chosen from NN3 Forecasting Competition are used. An average smoothing technique is assumed to complete the data missing in the dataset. The Hurst parameter estimated through wavelets is used to estimate the roughness of the real and forecasted series. The validation and horizon of the time series is presented by the 15 values ahead. The performance of the proposed filter shows that even a short dataset is incomplete, besides a linear smoothing technique employed; the prediction is almost fair by means of SMAPE index. Although the major result shows that the predictor system based on energy associated to series has an optimal performance from several chaotic time series, in particular, this method among other provides a good estimation when the short-term series are taken from one point observations.

ENSO response to external forcing in CMIP5 simulations of the last millennium

The El Niño–Southern Oscillation (ENSO) is dominant mode of interannual climate variability, but its response to external climate forcings remains uncertain. Past studies have limitations including the use of short datasets and the uncertainty contained in reconstructions or simulations of past ENSO variations. To improve our understanding on ENSO variations, it is important to examine its response to these forcings by using multi model simulations since they provide longer datasets and help improving the statistical significance of the results. In this study, Granger causality test is applied to investigate the influence of external forcings on ENSO by using past millennium simulations (period 850–1850CE) of Coupled Model Intercomparison Project Phase 5 (CMIP5) models. The results show robust influence of volcanic forcing to ENSO during preindustrial times of last millennium. The response of ENSO to solar forcing is more likely to be weak. Detection of GHGs variations signal in ENSO response is not significant. However, this study also indicates that there are uncertainties in the responses of ENSO to solar forcing and GHGs radiative forcing. The possibility of the true causal connection between Total Solar Irradiance (TSI) or GHGs radiative forcing and ENSO cannot be rejected at 95% significance level.

Reliable Analysis of Single-Unit Recordings from the Human Brain under Noisy Conditions: Tracking Neurons over Hours.

Recording extracellulary from neurons in the brains of animals in vivo is among the most established experimental techniques in neuroscience, and has recently become feasible in humans. Many interesting scientific questions can be addressed only when extracellular recordings last several hours, and when individual neurons are tracked throughout the entire recording. Such questions regard, for example, neuronal mechanisms of learning and memory consolidation, and the generation of epileptic seizures. Several difficulties have so far limited the use of extracellular multi-hour recordings in neuroscience: Datasets become huge, and data are necessarily noisy in clinical recording environments. No methods for spike sorting of such recordings have been available. Spike sorting refers to the process of identifying the contributions of several neurons to the signal recorded in one electrode. To overcome these difficulties, we developed Combinato: a complete data-analysis framework for spike sorting in noisy recordings lasting twelve hours or more. Our framework includes software for artifact rejection, automatic spike sorting, manual optimization, and efficient visualization of results. Our completely automatic framework excels at two tasks: It outperforms existing methods when tested on simulated and real data, and it enables researchers to analyze multi-hour recordings. We evaluated our methods on both short and multi-hour simulated datasets. To evaluate the performance of our methods in an actual neuroscientific experiment, we used data from from neurosurgical patients, recorded in order to identify visually responsive neurons in the medial temporal lobe. These neurons responded to the semantic content, rather than to visual features, of a given stimulus. To test our methods with multi-hour recordings, we made use of neurons in the human medial temporal lobe that respond selectively to the same stimulus in the evening and next morning.

Effects of adenosine and regadenoson on hemodynamics measured using cardiovascular magnetic resonance imaging

BackgroundAdenosine or regadenoson vasodilator stress cardiovascular magnetic resonance (CMR) is an effective non-invasive strategy for evaluating symptomatic coronary artery disease. Vasodilator injection typically precedes ventricular functional sequences to efficiently reduce overall scanning times, though the effects of vasodilators on CMR-derived ventricular volumes and function are unknown.MethodsWe prospectively enrolled 25 healthy subjects to undergo consecutive adenosine and regadenoson administration. Short axis CINE datasets were obtained on a 1.5 T scanner following adenosine (140mcg/kg/min IV for 6 min) and regadenoson (0.4 mg IV over 10 s) at baseline, immediately following administration, at 5 min intervals up to 15 min. Hemodynamic response, bi-ventricular volumes and ejection fractions were determined at each time point.ResultsPeak heart rate was observed early following administration of both adenosine and regadenoson. Heart rate returned to baseline by 10 min post-adenosine while remaining elevated at 15 min post-regadenoson (p = 0.0015). Left ventricular (LV) ejection fraction (LVEF) increased immediately following both vasodilators (p < 0.0001 for both) and returned to baseline following adenosine by 10 min (p = 0.8397). Conversely, LVEF following regadenoson remained increased at 10 min (p = 0.003) and 15 min (p = 0.0015) with a mean LVEF increase at 15 min of 4.2 ± 1.3%. Regadenoson resulted in a similar magnitude reduction in both LV end-diastolic volume index (LVEDVi) and LV end-systolic volume index (LVESVi) at 15 min whereas LVESVi resolved at 15 min following adenosine and LVEDVi remained below baseline values (p = 0.52).ConclusionsRegadenoson and adenosine have significant and prolonged impact on ventricular volumes and LVEF. In patients undergoing vasodilator stress CMR where ventricular volumes and LVEF are critical components to patient care, ventricular functional sequences should be performed prior to vasodilator use or consider the use of aminophylline in the setting of regadenoson. Additionally, heart rate resolution itself is not an effective surrogate for return of ventricular volumes and LVEF to baseline.

MetaProb: accurate metagenomic reads binning based on probabilistic sequence signatures.

Sequencing technologies allow the sequencing of microbial communities directly from the environment without prior culturing. Taxonomic analysis of microbial communities, a process referred to as binning, is one of the most challenging tasks when analyzing metagenomic reads data. The major problems are the lack of taxonomically related genomes in existing reference databases, the uneven abundance ratio of species and the limitations due to short read lengths and sequencing errors. MetaProb is a novel assembly-assisted tool for unsupervised metagenomic binning. The novelty of MetaProb derives from solving a few important problems: how to divide reads into groups of independent reads, so that k-mer frequencies are not overestimated; how to convert k-mer counts into probabilistic sequence signatures, that will correct for variable distribution of k-mers, and for unbalanced groups of reads, in order to produce better estimates of the underlying genome statistic; how to estimate the number of species in a dataset. We show that MetaProb is more accurate and efficient than other state-of-the-art tools in binning both short reads datasets (F-measure 0.87) and long reads datasets (F-measure 0.97) for various abundance ratios. Also, the estimation of the number of species is more accurate than MetaCluster. On a real human stool dataset MetaProb identifies the most predominant species, in line with previous human gut studies. https://bitbucket.org/samu661/metaprob cinzia.pizzi@dei.unipd.it or comin@dei.unipd.it Supplementary data are available at Bioinformatics online.

Minimum temperatures and frost at Grootfontein in the eastern Karoo, South Africa, over 98 years

Minimum temperatures and frost are ecologically and agriculturally important parameters as they can cause mortality in plants and livestock, shape the position of biome boundaries, influence the lifecycles of pests and diseases, and influence the availability and quality of livestock feed. This study examines minimum temperature data from 1916 to 2014 for a site on the Nama-Karoo/Grassland biome ecotone. Temperatures of at least −6 °C occur almost every year, while temperatures of below −10 °C occurred in 15 of the years on record. Temperature data did not exhibit a random pattern, but rather showed the period from 1935 to 1960 to be particularly cold in terms of absolute minimum temperatures, the probability of occurrence of frost and the duration of the dormant (frost) season. The results provide a platform for interpretation for ecological and agricultural processes that have been recorded in several long-term trials at Grootfontein. The patterns of variation in the data indicate that short (c. 50-year) temperature datasets may give misleading indications in trends in temperature, and for an empirically-based interpretation of climate change, long-term datasets of approximately a hundred years are needed.

Factors Affecting the Segmentation of the Heart Ventricles in Short Axis Cardiac Perfusion MRI Images

Cardiovascular diseases (CVDs) cause 31% of the death rate globally. Automatic accurate segmentation is needed for CVDs early detection. In this paper, we study the effect of the registration and initialization of the level set segmentation on the performance of extracting the heart ventricles for the short axis cardiac perfusion MRI images, as a result, we propose a modified workflow to automatically segment the ventricles by mitigating the levelset initial contour extraction in order to improve the segmentation results accuracy. In the registration experiments, the translational transformation was studied based on both the spatial and frequency domain. The frequency domain based registration is mainly established based on the phase correlation methodology. As for the segmentation experiments, the level set initialization was done through extracting the ventricles’ real shape from each slice. Though, the final contour of any frame will be used as the initial contour for the next frame. The second initialization strategy was based on defining the initial contour for each frame using the polar representation of the image. Two short axis view datasets of cardiac magnetic resonance (CMR) perfusion imaging were used in testing the proposed methods. Dice coefficient, sensitivity, specificity and Hausdorff distance have been used to evaluate and validate the segmentation results. The system workflow consists of five main modules: preprocessing, localization, initial contour extraction, registration, and segmentation. The segmentation accuracy for left and right ventricles improved from 72% to 77% and from 70% to 81% using the spatial domain based registration algorithm. The polar-based initialization strategy improves the segmentation accuracy from 77% to 81% and from 81% to 82% for the left and right ventricles respectively.

The Use of TRMM 3B42 Product for Drought Monitoring in Mexico

Drought has been a recurrent phenomenon in Mexico. For its assessment and monitoring, several studies have monitored meteorological droughts using standardized indices of precipitation deficits. Such conventional studies have mostly relied on rain gauge-based measurements, with the main limitation being the scarcity of rain gauge spatial coverage. This issue does not allow a robust spatial characterization of drought. A recent alternative for monitoring purposes can be found in satellite-based remote sensing of meteorological variables. The main objective of this study is to evaluate the standardized precipitation index (SPI) in Mexico during the period 1998 to 2013, using the Tropical Rainfall Measuring Mission (TRMM) satellite product 3B42. Results suggest that Mexico experienced the driest conditions during the great drought between 2011 and 2012; however, temporal variability in the SPI was found across different climatic regions. Nevertheless, a comparison of the SPI derived by TRMM against the rain gauge-based SPI computed by the official Mexican Drought Monitor showed low to medium correlation of the time series though both SPIs managed to capture the most relevant droughts at the national scale. We conclude that the TRMM product can properly monitor meteorological droughts despite its relative short dataset length (~15 years). Finally, we recommend an assimilation of rain gauge and satellite-based precipitation data to provide more robust estimates of meteorological drought severity.

Active Learning for Regression With Correlation Matching and Labeling Error Suppression

In this letter, we develop an active learning algorithm to optimize the selection of training data for robust linear regression. This algorithm selects training data based on the principle of correlation matching between the training dataset and the overall data pool. Considering the inevitable and potentially heavy human labeling errors, we model the probability of labeling errors based on the item response theory (IRT) and develop data screening techniques to control the error sparsity. Compressive sensing theory is then exploited for human labeling error suppression. This algorithm is robust even in the case of short training dataset with nonsparse labeling errors. Its performance is verified by simulations with both artificial data and real benchmark data. Experiments are also conducted to demonstrate the validity of the IRT-based human labeling error model and the superior performance of the algorithm in practical applications.

Three-factor models versus time series models: quantifying time-dependencies of interactions between stimuli in cell biology and psychobiology for short longitudinal data.

Signal integration determines cell fate on the cellular level, affects cognitive processes and affective responses on the behavioural level, and is likely to be involved in psychoneurobiological processes underlying mood disorders. Interactions between stimuli may subjected to time effects. Time-dependencies of interactions between stimuli typically lead to complex cell responses and complex responses on the behavioural level. We show that both three-factor models and time series models can be used to uncover such time-dependencies. However, we argue that for short longitudinal data the three factor modelling approach is more suitable. In order to illustrate both approaches, we re-analysed previously published short longitudinal data sets. We found that in human embryonic kidney 293 cells cells the interaction effect in the regulation of extracellular signal-regulated kinase (ERK) 1 signalling activation by insulin and epidermal growth factor is subjected to a time effect and dramatically decays at peak values of ERK activation. In contrast, we found that the interaction effect induced by hypoxia and tumour necrosis factor-alpha for the transcriptional activity of the human cyclo-oxygenase-2 promoter in HEK293 cells is time invariant at least in the first 12-h time window after stimulation. Furthermore, we applied the three-factor model to previously reported animal studies. In these studies, memory storage was found to be subjected to an interaction effect of the beta-adrenoceptor agonist clenbuterol and certain antagonists acting on the alpha-1-adrenoceptor / glucocorticoid-receptor system. Our model-based analysis suggests that only if the antagonist drug is administer in a critical time window, then the interaction effect is relevant.

Improving short text classification by learning vector representations of both words and hidden topics

This paper presents a general framework for short text classification by learning vector representations of both words and hidden topics together. We refer to a large-scale external data collection named ”corpus” which is topic consistent with short texts to be classified and then use the corpus to build topic model with Latent Dirichlet Allocation (LDA). For all the texts of the corpus and short texts, topics of words are viewed as new words and integrated into texts for data enriching. On the enriched corpus, we can learn vector representations of both words and topics. In this way, feature representations of short texts can be performed based on vectors of both words and topics for training and classification. On an open short text classification data set, learning vectors of both words and topics can significantly help reduce the classification error comparing with learning only word vectors. We also compared the proposed classification method with various baselines and experimental results justified the effectiveness of our word/topic vector representations.

Refined analyses suggest that recombination is a minor source of genomic diversity in Pseudomonas aeruginosa chronic cystic fibrosis infections.

Chronic bacterial airway infections in people with cystic fibrosis (CF) are often caused by Pseudomonas aeruginosa, typically showing high phenotypic diversity amongst co-isolates from the same sputum sample. Whilst adaptive evolution during chronic infections has been reported, the genetic mechanisms underlying the observed rapid within-population diversification are not well understood. Two recent conflicting reports described very high and low rates of homologous recombination in two closely related P. aeruginosa populations from the lungs of different chronically infected CF patients. To investigate the underlying cause of these contrasting observations, we combined the short read datasets from both studies and applied a new comparative analysis. We inferred low rates of recombination in both populations. The discrepancy in the findings of the two previous studies can be explained by differences in the application of variant calling techniques. Two novel algorithms were developed that filter false-positive variants. The first algorithm filters variants on the basis of ambiguity within duplications in the reference genome. The second omits probable false-positive variants at regions of non-homology between reference and sample caused by structural rearrangements. As gains and losses of prophage or genomic islands are frequent causes of chromosomal rearrangements within microbial populations, this filter has broad appeal for mitigating false-positive variant calls. Both algorithms are available in a Python package.

Forecasting medical waste generation using short and extra short datasets: Case study of Lithuania.

The aim of the study is to evaluate the performance of various mathematical modelling methods, while forecasting medical waste generation using Lithuania's annual medical waste data. Only recently has a hazardous waste collection system that includes medical waste been created and therefore the study access to gain large sets of relevant data for its research has been somewhat limited. According to data that was managed to be obtained, it was decided to develop three short and extra short datasets with 20, 10 and 6 observations. Spearman's correlation calculation showed that the influence of independent variables, such as visits at hospitals and other medical institutions, number of children in the region, number of beds in hospital and other medical institutions, average life expectancy and doctor's visits in that region are the most consistent and common in all three datasets. Tests on the performance of artificial neural networks, multiple linear regression, partial least squares, support vector machines and four non-parametric regression methods were conducted on the collected datasets. The best and most promising results were demonstrated by generalised additive (R(2) = 0.90455) in the regional data case, smoothing splines models (R(2) = 0.98584) in the long annual data case and multilayer feedforward artificial neural networks in the short annual data case (R(2) = 0.61103).

Detection of Coupling in Short Physiological Series by a Joint Distribution Entropy Method.

In this study, we developed a joint distribution entropy (JDistEn) method to robustly estimate the coupling in short physiological series. The JDistEn method is derived from a joint distance matrix which is constructed from a combination of the distance matrix corresponding to each individual data channel using a geometric mean calculation. A coupled Rössler system and a coupled dual-kinetics neural mass model were used to examine how well JDistEn performed, specifically, its sensitivity for detecting weak coupling, its consistency in gauging coupling strength, and its reliability in processing input of decreased data length. Performance of JDistEn in estimating physiological coupling was further examined with bivariate electroencephalography data from rats and RR interval and diastolic time interval series from human beings. Cross-sample entropy (XSampEn), cross-conditional entropy (XCE), and Shannon entropy of diagonal lines in the joint recurrence plots (JENT) were applied for purposes of comparison. Simulation results suggest that JDistEn showed markedly higher sensitivity than XSampEn, XCE, and JENT for dynamics in weak coupling, although as the simulation models were more intensively coupled, JDistEn performance was comparable to the three others. In addition, this improved sensitivity was much more pronounced for short datasets. Experimental results further confirmed that JDistEn outperformed XSampEn, XCE, and JENT for detecting weak coupling, especially for short physiological data. This study suggested that our proposed JDistEn could be useful for continuous and even real-time coupling analysis for physiological signals in clinical practice.

ESTIMATION OF RAINFALL BY NEURAL NETWORK OVER A NEOTROPICAL REGION (ESTIMATIVA DE CHUVAS PELA REDE NEURAL SOBRE A REGIÃO NEOTROPICAL)

Rainfall is the key element in regional water balance, and has direct influence over economic activity. Quantifying rainfall at spatial and temporal scales in regions where meteorological stations are scarce is important for agriculture, natural resource management and land-atmosphere interactions science. Thus, we evaluated neural network performance for rainfall estimates over Mato Grosso State located in the Brazilian Midwest region. A dataset of 12 meteorological stations was used to train the neural network, and then was run to perform estimates, which allowed comparing with TRMM satellite estimates. Rainfall estimates were performed by neural network as a function of latitude and longitude for model 1 (NN1), and latitude, longitude, and altitude for model 2 (NN2). In general, rainfall values were higher overestimated by NN1 in January (18.7%), and September (1,501%) than by NN2 in January (6.8%), and September (0%). While NN2 had rainfall pattern closer to TRMM both in January and in September, NN1 rainfall estimates captured a small amount of the rainfall pattern over Mato Grosso state. There was no higher difference between NN1 and NN2 rainfall estimates in January. On the other hand, rainfall estimates had better performance in NN2 than NN1 in September. Anyway, rainfall estimates were worst in January than in September both in NN1 and in NN2. Bad rainfall estimates using neural network in Mato Grosso state were due to (i) a short temporal dataset, (ii) few stations with poor spatial variability, (iii) few auxiliary variables to build neural network, which could better capture the rain phenomenon over Neotropical regions. The next step will be to analyze the rainfall and other climatic estimates performed by neural network for the whole year for several years on the Central-West Region of Brazil including other auxiliary variables besides latitude, longitude, and altitude.