Ethnologue: Languages of the World

Multivariate LSTM-FCNs for time series classification.

Dataset augmentation, the practice of applying a wide array of domain-specific transformations to synthetically expand a training set, is a standard tool in supervised learning. While effective in tasks such as visual recognition, the set of transformations must be carefully designed, implemented, and tested for every new domain, limiting its re-use and generality. In this paper, we adopt a simpler, domain-agnostic approach to dataset augmentation. We start with existing data points and apply simple transformations such as adding noise, interpolating, or extrapolating between them. Our main insight is to perform the transformation not in input space, but in a learned feature space. A re-kindling of interest in unsupervised representation learning makes this technique timely and more effective. It is a simple proposal, but to-date one that has not been tested empirically. Working in the space of context vectors generated by sequence-to-sequence models, we demonstrate a technique that is effective for both static and sequential data.

Dataset Augmentation in Feature Space

In 2002, the UCR time series classification archive was first released with sixteen datasets. It gradually expanded, until 2015 when it increased in size from 45 datasets to 85 datasets. In October 2018 more datasets were added, bringing the total to 128. The new archive contains a wide range of problems, including variable length series, but it still only contains univariate time series classification problems. One of the motivations for introducing the archive was to encourage researchers to perform a more rigorous evaluation of newly proposed time series classification (TSC) algorithms. It has worked: most recent research into TSC uses all 85 datasets to evaluate algorithmic advances. Research into multivariate time series classification, where more than one series are associated with each class label, is in a position where univariate TSC research was a decade ago. Algorithms are evaluated using very few datasets and claims of improvement are not based on statistical comparisons. We aim to address this problem by forming the first iteration of the MTSC archive, to be hosted at the website this http URL. Like the univariate archive, this formulation was a collaborative effort between researchers at the University of East Anglia (UEA) and the University of California, Riverside (UCR). The 2018 vintage consists of 30 datasets with a wide range of cases, dimensions and series lengths. For this first iteration of the archive we format all data to be of equal length, include no series with missing data and provide train/test splits.

The UEA multivariate time series classification archive, 2018.

Multivariate time series (MTS) classification has gained importance with the increase in the number of temporal datasets in different domains (such as medicine, finance, multimedia, etc.). Similarity-based approaches, such as nearest-neighbor classifiers, are often used for univariate time series, but MTS are characterized not only by individual attributes, but also by their relationships. Here we provide a classifier based on a new symbolic representation for MTS (denoted as SMTS) with several important elements. SMTS considers all attributes of MTS simultaneously, rather than separately, to extract information contained in the relationships. Symbols are learned from a supervised algorithm that does not require pre-defined intervals, nor features. An elementary representation is used that consists of the time index, and the values (and first differences for numerical attributes) of the individual time series as columns. That is, there is essentially no feature extraction (aside from first differences) and the local series values are fused to time position through the time index. The initial representation of raw data is quite simple conceptually and operationally. Still, a tree-based ensemble can detect interactions in the space of the time index and time values and this is exploited to generate a high-dimensional codebook from the terminal nodes of the trees. Because the time index is included as an attribute, each MTS is learned to be segmented by time, or by the value of one of its attributes. The codebook is processed with a second ensemble where now implicit feature selection is exploited to handle the high-dimensional input. The constituent properties produce a distinctly different algorithm. Moreover, MTS with nominal and missing values are handled efficiently with tree learners. Experiments demonstrate the effectiveness of the proposed approach in terms of accuracy and computation times in a large collection multivariate (and univariate) datasets.

Learning a symbolic representation for multivariate time series classification

This paper introduces a fast learning method for a graphical probabilistic model for discrete speech recognition based on spoken Arabic digit recognition by means of a new proposed spanning tree structure that takes advantage of the temporal nature of speech signal. The experimental results obtained on a spoken Arabic digit dataset confirmed that for the same rate of recognition the proposed method, in terms of time computation is much faster than the state of art algorithm that use the maximum weight spanning tree (MWST).

Improved tree model for arabic speech recognition

In this work we propose a novel method for automatic discrete speech recognition composed from two steps. In a first step, discrete speech features are extracted by means of Mel Frequency Cepstral Coefficients (MFCCs) followed by vector quantization (VQ). Then in a second step, the obtained features are fed to a Tree distribution classifier which provides the class-label associated with each feature by approximating the true class probability by means of an optimal spanning tree model. The experimental results obtained on a spoken Arabic digit dataset confirmed the promising capabilities of the proposed approach.

Tree distribution classifier for automatic spoken Arabic digit recognition

This paper proposes a logistic regression approach paired with term and inverse document frequency (TF*IDF) for Arabic sentiment classification on services’ reviews in Lebanon country. Reviews are about public services, including hotels, restaurants, shops, and others. We collected manually from Google reviews and Zomato, which have reached to 3916 reviews. Experiments show three core findings: 1) The classifier is confident when used to predict positive reviews. 2) The model is biased on predicting reviews with negative sentiment. Finally, the low percentage of negative reviews in the corpus contributes to the diffidence of logistic regression model.

Sentiment Classifier: Logistic Regression for Arabic Services’ Reviews in Lebanon

Mel Frequency Cepstral Coefficients (MFCCs) are the most popularly used speech features in many speech and speaker recognition applications. In this paper, we study the effect of the second-order derivatives of MFCC on the recognition of the Spoken Arabic digits. The system was developed using the Hidden Markov Models (HMMs) and Tree distribution approximation model. Experimentally it has been shown that, the second-order derivatives of MFCC parameters compared to the MFCC yield improved rates of 4.60% for CHMM. We were able to reach an overall recognition accuracy of 98.41%, which is satisfactory compared to previous work on spoken Arabic digits speech recognition.

The second-order derivatives of MFCC for improving spoken Arabic digits recognition using Tree distributions approximation model and HMMs

Breast cancer is the fifth most common cause of cancer death among women worldwide, even on Algeria that known about 12,000 new cases every year. Texture description has been a great interest in pattern recognition methods for looking deeper into features images, In this paper, we investigate the capability of the Local Binary Pattern texture and deep learning method for automated breast tumor images classification to be an efficient element for Computer aided diagnosis (CAD) system, where the extraction of meaningful information from the input image do not require features extractors. We have proposed a Convolution Neural Network (CNN) architecture based on LBP images as input after we compared their classification results by a standard CNN based on origin images as input. A 190-segmented image from (DDSM) database will be used for testing the proposed approach. Experimental results of the classification (benign or malignant tumor) gave better results than the standard CNN approach with an overall accuracy about 96.32 %.

Nacereddine Hammami

Papers

Improved tree model for arabic speech recognition

Tree distribution classifier for automatic spoken Arabic digit recognition

Sentiment Classifier: Logistic Regression for Arabic Services’ Reviews in Lebanon

The second-order derivatives of MFCC for improving spoken Arabic digits recognition using Tree distributions approximation model and HMMs

Automated Breast Tumor Diagnosis Using Local Binary Patterns (LBP) Based on Deep Learning Classification