Abstract: Over-parameterized neural networks have attracted a
great deal of attention in recent deep learning theory research,
as they challenge the classic perspective of over-fitting when
the model has excessive parameters and have gained empirical
success in various settings. While a number of theoretical works
have been presented to demystify properties of such models, the
convergence properties of such models are still far from being
thoroughly understood. In this work, we study the convergence
properties of training two-hidden-layer partially over-parameterized
fully connected networks with the Rectified Linear Unit activation via
gradient descent. To our knowledge, this is the first theoretical work
to understand convergence properties of deep over-parameterized
networks without the equally-wide-hidden-layer assumption and
other unrealistic assumptions. We provide a probabilistic lower bound
of the widths of hidden layers and proved linear convergence rate of
gradient descent. We also conducted experiments on synthetic and
real-world datasets to validate our theory.
Abstract: In Algeria agricultural irrigation is the primary water consuming sector followed by the domestic and industrial sectors. Economic development in the last decade has weighed heavily on water resources which are relatively limited and gradually decreasing to the detriment of agriculture. The research presented in this paper focuses on the optimization of irrigation water demand. Dynamic Programming-Neural Network (DPNN) method is applied to investigate reservoir optimization. The optimal operation rule is formulated to minimize the gap between water release and water irrigation demand. As a case study, Foum El-Gherza dam’s reservoir system in south of Algeria has been selected to examine our proposed optimization model. The application of DPNN method allowed increasing the satisfaction rate (SR) from 12.32% to 55%. In addition, the operation rule generated showed more reliable and resilience operation for the examined case study.
Abstract: In the last few years, Automatic Number Plate Recognition (ANPR) systems have become widely used in the safety, the security, and the commercial aspects. Forethought, several methods and techniques are computing to achieve the better levels in terms of accuracy and real time execution. This paper proposed a computer vision algorithm of Number Plate Localization (NPL) and Characters Segmentation (CS). In addition, it proposed an improved method in Optical Character Recognition (OCR) based on Deep Learning (DL) techniques. In order to identify the number of detected plate after NPL and CS steps, the Convolutional Neural Network (CNN) algorithm is proposed. A DL model is developed using four convolution layers, two layers of Maxpooling, and six layers of fully connected. The model was trained by number image database on the Jetson TX2 NVIDIA target. The accuracy result has achieved 95.84%.
Abstract: Learning from very big datasets is a significant problem for most present data mining and machine learning algorithms. MicroRNA (miRNA) is one of the important big genomic and non-coding datasets presenting the genome sequences. In this paper, a hybrid method for the classification of the miRNA data is proposed. Due to the variety of cancers and high number of genes, analyzing the miRNA dataset has been a challenging problem for researchers. The number of features corresponding to the number of samples is high and the data suffer from being imbalanced. The feature selection method has been used to select features having more ability to distinguish classes and eliminating obscures features. Afterward, a Convolutional Neural Network (CNN) classifier for classification of cancer types is utilized, which employs a Genetic Algorithm to highlight optimized hyper-parameters of CNN. In order to make the process of classification by CNN faster, Graphics Processing Unit (GPU) is recommended for calculating the mathematic equation in a parallel way. The proposed method is tested on a real-world dataset with 8,129 patients, 29 different types of tumors, and 1,046 miRNA biomarkers, taken from The Cancer Genome Atlas (TCGA) database.
Abstract: The more an educational system knows about a learner, the more personalised interaction it can provide, which leads to better learning. However, asking a learner directly is potentially disruptive, and often ignored by learners. Especially in the booming realm of MOOC Massive Online Learning platforms, only a very low percentage of users disclose demographic information about themselves. Thus, in this paper, we aim to predict learners’ demographic characteristics, by proposing an approach using linguistically motivated Deep Learning Architectures for Learner Profiling, particularly targeting gender prediction on a FutureLearn MOOC platform. Additionally, we tackle here the difficult problem of predicting the gender of learners based on their comments only – which are often available across MOOCs. The most common current approaches to text classification use the Long Short-Term Memory (LSTM) model, considering sentences as sequences. However, human language also has structures. In this research, rather than considering sentences as plain sequences, we hypothesise that higher semantic - and syntactic level sentence processing based on linguistics will render a richer representation. We thus evaluate, the traditional LSTM versus other bleeding edge models, which take into account syntactic structure, such as tree-structured LSTM, Stack-augmented Parser-Interpreter Neural Network (SPINN) and the Structure-Aware Tag Augmented model (SATA). Additionally, we explore using different word-level encoding functions. We have implemented these methods on Our MOOC dataset, which is the most performant one comparing with a public dataset on sentiment analysis that is further used as a cross-examining for the models' results.
Abstract: Uncontrolled growth of abnormal cells in the lung in the form of tumor can be either benign (non-cancerous) or malignant (cancerous). Patients with Lung Cancer (LC) have an average of five years life span expectancy provided diagnosis, detection and prediction, which reduces many treatment options to risk of invasive surgery increasing survival rate. Computed Tomography (CT), Positron Emission Tomography (PET), and Magnetic Resonance Imaging (MRI) for earlier detection of cancer are common. Gaussian filter along with median filter used for smoothing and noise removal, Histogram Equalization (HE) for image enhancement gives the best results without inviting further opinions. Lung cavities are extracted and the background portion other than two lung cavities is completely removed with right and left lungs segmented separately. Region properties measurements area, perimeter, diameter, centroid and eccentricity measured for the tumor segmented image, while texture is characterized by Gray-Level Co-occurrence Matrix (GLCM) functions, feature extraction provides Region of Interest (ROI) given as input to classifier. Two levels of classifications, K-Nearest Neighbor (KNN) is used for determining patient condition as normal or abnormal, while Artificial Neural Networks (ANN) is used for identifying the cancer stage is employed. Discrete Wavelet Transform (DWT) algorithm is used for the main feature extraction leading to best efficiency. The developed technology finds encouraging results for real time information and on line detection for future research.
Abstract: This article contains information from our investigation in the field of voice recognition. For this purpose, we created a voice database that contains different phrases in two languages, English and Spanish, for men and women. As a classifier, the LIRA (Limited Receptive Area) grayscale neural classifier was selected. The LIRA grayscale neural classifier was developed for image recognition tasks and demonstrated good results. Therefore, we decided to develop a recognition system using this classifier for voice recognition. From a specific set of speakers, we can recognize the speaker’s voice. For this purpose, the system uses spectrograms of the voice signals as input to the system, extracts the characteristics and identifies the speaker. The results are described and analyzed in this article. The classifier can be used for speaker identification in security system or smart buildings for different types of intelligent devices.
Abstract: The shift towards decision making (DM) based on artificial intelligence (AI) techniques will change the way in which consumer markets and our societies function. Through AI, predictive analytics is being used by businesses to identify these patterns and major trends with the objective to improve the DM and influence future business outcomes. This paper proposes an Artificial Neural Network (ANN) approach to predict the success of telemarketing calls for selling bank long-term deposits. To validate the proposed model, we uses the bank marketing data of 41188 phone calls. The ANN attains 98.93% of accuracy which outperforms other conventional classifiers and confirms that it is credible and valuable approach for telemarketing campaign managers.
Abstract: Certain systems can function well only if they recognize the sound environment as humans do. In this research, we focus on sound classification by adopting a convolutional neural network and aim to develop a method that automatically classifies various environmental sounds. Although the neural network is a powerful technique, the performance depends on the type of input data. Therefore, we propose an approach via a slice bispectrogram, which is a third-order spectrogram and is a slice version of the amplitude for the short-time bispectrum. This paper explains the slice bispectrogram and discusses the effectiveness of the derived method by evaluating the experimental results using the ESC‑50 sound dataset. As a result, the proposed scheme gives high accuracy and stability. Furthermore, some relationship between the accuracy and non-Gaussianity of sound signals was confirmed.
Abstract: Speech to text in Malay language is a system that converts Malay speech into text. The Malay language recognition system is still limited, thus, this paper aims to investigate the performance of ten Malay words obtained from the online Malay news. The methodology consists of three stages, which are preprocessing, feature extraction, and speech classification. In preprocessing stage, the speech samples are filtered using pre emphasis. After that, feature extraction method is applied to the samples using Mel Frequency Cepstrum Coefficient (MFCC). Lastly, speech classification is performed using Feedforward Neural Network (FFNN). The accuracy of the classification is further investigated based on the hidden layer size. From experimentation, the classifier with 40 hidden neurons shows the highest classification rate which is 94%.
Abstract: The Bacterial Foraging Optimization (BFO) algorithm is inspired by the behavior of bacteria such as Escherichia coli or Myxococcus xanthus when searching for food, more precisely the chemotaxis behavior. Bacteria perceive chemical gradients in the environment, such as nutrients, and also other individual bacteria, and move toward or in the opposite direction to those signals. The application example considered as a case study consists in establishing the dependency between the reaction yield of hydrogels based on polyacrylamide and the working conditions such as time, temperature, monomer, initiator, crosslinking agent and inclusion polymer concentrations, as well as type of the polymer added. This process is modeled with a neural network which is included in an optimization procedure based on BFO. An experimental study of BFO parameters is performed. The results show that the algorithm is quite robust and can obtain good results for diverse combinations of parameter values.
Abstract: This paper aims at bringing a scientific contribution to the cardiac arrhythmia biomedical diagnosis systems; more precisely to the study of the amelioration of cardiac arrhythmia classification performance using artificial neural network, adaptive neuro-fuzzy and fuzzy inference systems classifiers. The purpose of this amelioration is to enable cardiologists to make reliable diagnosis through automatic cardiac arrhythmia analyzes and classifications based on high confidence classifiers. In this study, six classes of the most commonly encountered arrhythmias are considered: the Right Bundle Branch Block, the Left Bundle Branch Block, the Ventricular Extrasystole, the Auricular Extrasystole, the Atrial Fibrillation and the Normal Cardiac rate beat. From the electrocardiogram (ECG) extracted parameters, we constructed a matrix (360x360) serving as an input data sample for the classifiers based on neural networks and a matrix (1x6) for the classifier based on fuzzy logic. By varying three parameters (the quality of the neural network learning, the data size and the quality of the input parameters) the automatic classification permitted us to obtain the following performances: in terms of correct classification rate, 83.6% was obtained using the fuzzy logic based classifier, 99.7% using the neural network based classifier and 99.8% for the adaptive neuro-fuzzy based classifier. These results are based on signals containing at least 360 cardiac cycles. Based on the comparative analysis of the aforementioned three arrhythmia classifiers, the classifiers based on neural networks exhibit a better performance.
Abstract: Prediction of wall shear stress in a rectangular channel, with non-homogeneous roughness distribution, was studied. Estimation of shear stress is an important subject in hydraulic engineering, since it affects the flow structure directly. In this study, the Genetic Algorithm Artificial (GAA) neural network is introduced as a hybrid methodology of the Artificial Neural Network (ANN) and modified Genetic Algorithm (GA) combination. This GAA method was employed to predict the wall shear stress. Various input combinations and transfer functions were considered to find the most appropriate GAA model. The results show that the proposed GAA method could predict the wall shear stress of open channels with high accuracy, by Root Mean Square Error (RMSE) of 0.064 in the test dataset. Thus, using GAA provides an accurate and practical simple-to-use equation.
Abstract: Yi is an ethnic group mainly living in mainland China, with its own spoken and written language systems, after development of thousands of years. Ancient Yi is one of the six ancient languages in the world, which keeps a record of the history of the Yi people and offers documents valuable for research into human civilization. Recognition of the characters in ancient Yi helps to transform the documents into an electronic form, making their storage and spreading convenient. Due to historical and regional limitations, research on recognition of ancient characters is still inadequate. Thus, deep learning technology was applied to the recognition of such characters. Five models were developed on the basis of the four-layer convolutional neural network (CNN). Alpha-Beta divergence was taken as a penalty term to re-encode output neurons of the five models. Two fully connected layers fulfilled the compression of the features. Finally, at the softmax layer, the orthographic features of ancient Yi characters were re-evaluated, their probability distributions were obtained, and characters with features of the highest probability were recognized. Tests conducted show that the method has achieved higher precision compared with the traditional CNN model for handwriting recognition of the ancient Yi.
Abstract: The neural network quantization is highly desired
procedure to perform before running neural networks on mobile
devices. Quantization without fine-tuning leads to accuracy drop of
the model, whereas commonly used training with quantization is done
on the full set of the labeled data and therefore is both time- and
resource-consuming. Real life applications require simplification and
acceleration of quantization procedure that will maintain accuracy of
full-precision neural network, especially for modern mobile neural
network architectures like Mobilenet-v1, MobileNet-v2 and MNAS. Here we present a method to significantly optimize training with
quantization procedure by introducing the trained scale factors for
discretization thresholds that are separate for each filter. Using the
proposed technique, we quantize the modern mobile architectures of
neural networks with the set of train data of only ∼ 10% of the
total ImageNet 2012 sample. Such reduction of train dataset size and
small number of trainable parameters allow to fine-tune the network
for several hours while maintaining the high accuracy of quantized
model (accuracy drop was less than 0.5%). Ready-for-use models and
code are available in the GitHub repository.
Abstract: 2016 has become the year of the Artificial Intelligence explosion. AI technologies are getting more and more matured that most world well-known tech giants are making large investment to increase the capabilities in AI. Machine learning is the science of getting computers to act without being explicitly programmed, and deep learning is a subset of machine learning that uses deep neural network to train a machine to learn features directly from data. Deep learning realizes many machine learning applications which expand the field of AI. At the present time, deep learning frameworks have been widely deployed on servers for deep learning applications in both academia and industry. In training deep neural networks, there are many standard processes or algorithms, but the performance of different frameworks might be different. In this paper we evaluate the running performance of two state-of-the-art distributed deep learning frameworks that are running training calculation in parallel over multi GPU and multi nodes in our cloud environment. We evaluate the training performance of the frameworks with ResNet-50 convolutional neural network, and we analyze what factors that result in the performance among both distributed frameworks as well. Through the experimental analysis, we identify the overheads which could be further optimized. The main contribution is that the evaluation results provide further optimization directions in both performance tuning and algorithmic design.
Abstract: Seizure is the main factor that affects the quality of life of epileptic patients. The diagnosis of epilepsy, and hence the identification of epileptogenic zone, is commonly made by using continuous Electroencephalogram (EEG) signal monitoring. Seizure identification on EEG signals is made manually by epileptologists and this process is usually very long and error prone. The aim of this paper is to describe an automated method able to detect seizures in EEG signals, using knowledge discovery in database process and data mining methods and algorithms, which can support physicians during the seizure detection process. Our detection method is based on Artificial Neural Network classifier, trained by applying the multilayer perceptron algorithm, and by using a software application, called Training Builder that has been developed for the massive extraction of features from EEG signals. This tool is able to cover all the data preparation steps ranging from signal processing to data analysis techniques, including the sliding window paradigm, the dimensionality reduction algorithms, information theory, and feature selection measures. The final model shows excellent performances, reaching an accuracy of over 99% during tests on data of a single patient retrieved from a publicly available EEG dataset.
Abstract: A major problem that affects the quality control of polymer in the industrial polymerization is the lack of suitable on-line measurement tools to evaluate the properties of the polymer such as melt and density indices. Controlling the polymerization in ordinary method is performed manually by taking samples, measuring the quality of polymer in the lab and registry of results. This method is highly time consuming and leads to producing large number of incompatible products. An online application for estimating melt index and density proposed in this study is a neural network based on the input-output data of the polyethylene production plant. Temperature, the level of reactors' bed, the intensity of ethylene mass flow, hydrogen and butene-1, the molar concentration of ethylene, hydrogen and butene-1 are used for the process to establish the neural model. The neural network is taught based on the actual operational data and back-propagation and Levenberg-Marquart techniques. The simulated results indicate that the neural network process model established with three layers (one hidden layer) for forecasting the density and the four layers for the melt index is able to successfully predict those quality properties.
Abstract: In this work, a training algorithm for probabilistic neural networks (PNN) is presented. The algorithm addresses one of the major drawbacks of PNN, which is the size of the hidden layer in the network. By using a cross-validation training algorithm, the number of hidden neurons is shrunk to a smaller number consisting of the most representative samples of the training set. This is done without affecting the overall architecture of the network. Performance of the network is compared against performance of standard PNN for different databases from the UCI database repository. Results show an important gain in network size and performance.
Abstract: Deep learning structure is a branch of machine learning science and greet achievement in research and applications. Cellular neural networks are regarded as array of nonlinear analog processors called cells connected in a way allowing parallel computations. The paper discusses how to use deep learning structure for representing neural cellular automata model. The proposed learning technique in cellular automata model will be examined from structure of deep learning. A deep automata neural cellular system modifies each neuron based on the behavior of the individual and its decision as a result of multi-level deep structure learning. The paper will present the architecture of the model and the results of simulation of approach are given. Results from the implementation enrich deep neural cellular automata system and shed a light on concept formulation of the model and the learning in it.