Sharif Digital Repository

Cerebrovascular Attack has been ranked the second or third of top 10 death causes in Taiwan. It has caused about 13,000 deaths every year since 1986. Once Cerebrovascular Attack (CVA) occurs, it not only leads to the huge cost of medical care, but even death. All developed countries in the world put CVA prevention and treatment in high priority. However, it is necessary to build a detective model to enhance the accurate diagnosis of CVA. From this detective model, CVA classification rules were extracted and used to improve the diagnosis and detection of CVA. This study acquired 2449 valid samples from this CVA prevention and treatment program, and adopted three classification algorithms,... 

Hyperspectral unmixing (HU) is a method used to estimate the fractional abundances corresponding to endmembers in each of the mixed pixels in the hyperspectral remote sensing image. In recent times, deep learning has been recognized as an effective technique for hyperspectral image classification. In this thesis, an end-to-end HU method is proposed based on the convolutional neural network (CNN) and multi-layer perceptron (MLP). which consists of two steps: the first stage extracts features from the input data along with the inverse learning of the spectral library matrix in the hyperspectral image where columns represent the pure spectral of endmembers and The second stage is to estimate... 

Air pollution caused by industrialization is the problem which adversely affects human life. Among air pollutants suspended particles, especially particles smaller than 10 microns (PM10), for their high concentration in air in large cities are the major index as air pollutant. Due to their small size, PM10 can penetrate into the aspiration organs causing harmful effects. The objective of this work is to develop an Artificial Neural Network (ANN) model for prediction of short-term concentration of PM10 in the city of Tehran. Complex mechanism of reactions, numerous types of pollutant materials produced from transportation and industrial activities, variety of sources, difficulties in data... 

Determining the porosity, water saturation and permeability values in the reservoir rock, are major steps in the petroleum engineering and formation evaluation. Today in the oil industry, these parameters are obtained using Helium gas injection technique on core (Plug) samples. However, the coring operation is difficult and costly. In addition, there isn't the possibility of coring in some of wells, such as horizontal wells. In fact, the present research is the modelling of artificial neural networks to estimate permeability in one of the Iranian oil field reservoirs by using oil well logging data. In this research, MLP neural network has been used and for network training, the evolutionary... 

The purpose of the present study is to develop neural network estimation method for hybrid or panel data which have a combination of two cross-sectional and time series structures and because of the features of both structures, their use in different sciences offers many advantages; and the analytical methods for this data structure are also different from other one-dimensional structures, so different and specific regression models are presented for this data structure. However, in the artificial neural network method, modeling the development for this data structure is neglected, so in the present study, using the concepts of panel regression methods and their application to the... 

Predicting oil prices, especially in exporting countries, will help governments in the policy-making process by obtaining a reliable estimate of oil revenues. The existence of a complex mechanism governing the process of oil price formation has reduced the efficiency of linear models in forecasting and led researchers to use nonlinear intelligent systems to predict oil prices. In this study, after a detailed study of the structure of artificial neural network, two models of neural network GMDH and MLP and ARIMA method have been used to predict oil price. There are important factors in the prediction process with neural networks, and if all these factors are selected correctly; One can expect... 

In this study, in order to enhance the accuracy of tracking repetitive maneuvers in Unmanned Aerial Vehicles (UAVs), an educable control scheme is proposed. At the outset, the controller is designed based on the sliding mode control (SMC) technique. In addition, the offline PD-type memory-based iterative learning control (ILC) is used along with SMC. In ILC scheme, the error of states is saved during the maneuvers that will be used in the subsequent iteration. Also, in order to increase flexibility of the new control structure, ILC-SMC, a multilayer perceptron has been developed. This network is designed to extend the control signal, generated by ILC, to similar maneuvers. The presented... 

Selecting a suitable material is crucial for repairing the old concrete structures and joining precast panels of bridges, especially the bond strength between the substrate concrete and the overlay material. In this regard, this research focused on high-performance alkali-activated slag concrete (HPASC) as a new concrete used as an overlay on ordinary Portland cement concrete (OPCC) as a block of old concrete. Approximately four hundred bi-surface shear (BSS) tests were performed to evaluate the interface properties of OPCC and HPASC. HPASC specimens were designed with different NaOH molarity, silica fume (SF) content, steel fiber content, age of repair material, and proportion of grooved... 

Data mining is the process of extracting information from large data sets using algorithms and methods derived from the field of statistics, machine learning and database management systems. Data mining, popularly known as knowledge discovery in big data, enables companies and organizations to make informed decisions by collecting, aggregating, analyzing and accessing company data. The pharmaceutical industry is one of the most important levels of the drug supply chain, which has a significant impact on the healthcare sector of any society. In this context data mining can be used in various procedure such as discovery of a new medicine, sequential registration of clinical trials, combining... 

Flash calculations using equations of state are the basis of compositional simulators and inaccuracy in these calculations leads to errors in some time steps or even complete failure of a simulation. One of the problems of this approach is the use of iterative loops, which leads to a high computational cost. The purpose of this research is to use deep machine learning methods, namely multi-layer perceptrons and convolutional neural networks to solve this problem. At first, we defined two 5-component and 10-component hydrocarbon fluid models and used one of the conventional methods of flash calculations (Modified Successive Substitution method) in order to generate the data required for... 

This research investigates the break-up physics and spray characteristics of flow-blurring spray by implementing machine learning on numerical and experimental datasets. To do so, five crucial parameters of atomization, including SMD, axial and radial velocity components, penetration length, break-up length, and spray angle, are selected to be studied. Firstly, size and velocity distribution datasets are gathered using available experimental papers. Prior to modeling by Multi-Layer Perceptron neural networks, the datasets were pre-processed in terms of the existence of multi-value and outlier instances. Secondly, the physics of mixing flow inside the injection system was numerically... 

Optical character recognition (OCR) method has been used in converting printed text into editable text. OCR is very usefuland popular method in various applications. Accuracy of OCRn can be dependent on text preprocessing and segmentation algorithms. Sometimes it is difficult to retrieve text from the image because of different size, style, orientation, complex background of image etc. and Deep learning is a branch of machine learning based on a set of algorithms that attempt to model high-level abstractions in data by using multiple processing layers, with complex structures or otherwise, composed of multiple non-linear transformations and a deep belief network (DBN) is a probabilistic ,... 

Forex prediction is one of the inresting topics for researchers and traders. They are always trying to improve their accuracy. Recently machine learning technichs especially deep neural networks has shown great success and high accuracy for market prediction.In this paper we propose several multi-input deep neural networks to predict the direction and amount of change for euro/usd rate. Our models are based on scalping strategy fitting high frequency trading. Our first model is based on deep Long Short Term Memory layers which is our best model. Our second model is based on deep 1D- Convolutional layers. The last model is our simplest model which is based on MultiLayer Perceptron. Also we... 

Examining and improving driving behavior leads to a reduction in road accidents. Driver behavior is generally divided into two categories: aggressive and safe. Driving culture can be enhanced by identifying aggressive behavior and informing drivers directly. This information is used to assign safe drivers to the missions of transport fleet management companies and organizations. Also, this information is used by insurance companies or traffic police to apply discounts or fines. In any case, the detection and notification of aggressive behaviors would reduce accidents and improve the lives and mental safety of passengers and drivers. The detection of driving events is an introduction to the... 

Cardiovascular diseases (CVDs) have ranked first cause of deaths globally. In 2016, about 17.7 million people died from CVDs representing 31% of all world deaths. So, early intelligent detection of cardiovascular disease could help to save many lives in worldwide. There are several methods to analyze heart activity and to detect any abnormalities including Electrocardiogram, Stress test, Echocardiography, cardiac catheterization and coronary angiography.Among all methods, Electrocardiogram (ECG) is the most common and convenient type where it measures heart electrical activity and records it as a series of pulses. Analyzing these pulses would provide useful information about normal and... 

This effort attempts to study the change point problem in the area of non-linear profiles. Two methods for estimating the time of a single step change is proposed. In the first method a model consisting of two networks which is based on artificial neural networks is proposed. These networks are different only in their training data. One network is trained for ascending segments of the profile and the other is trained for descending segments of the profile. In the second method the maximum likelihood estimator (MLE) of the single step change is analyzed. Due to the complexity of estimating the parameters of the non-linear model by MLE, this estimator is based on the difference between the...