Soheil Mehralian | K.N.Toosi University of Technology (original) (raw)
Uploads
Papers by Soheil Mehralian
2020 8th Iranian Joint Congress on Fuzzy and intelligent Systems (CFIS), 2020
This paper presents the system identification based on a flexible deep neural network for a seven... more This paper presents the system identification based on a flexible deep neural network for a seven degree of freedom(7DOF), a full-car active suspension system that is multi-input and multi-output. The proposed flexible deep neural network, according to input and output data, obtained three layers of flexible auto-encoder. The flexible name was chosen for the learnable activation function parameter in the activation layers. This view permits every neuron to adjust its activation function and adapt the neuron to boost performance. Here flexible tanh activation function introduced, which causes better performance with the same neurons in the hidden layer. The comparison shows the identification error between flexible deep neural network and classical deep neural network. This adaptation, of course, provides prediction improvement.
Web Application Firewall (WAF) is known as one of the Intrusion Detection System (IDS) solutions ... more Web Application Firewall (WAF) is known as one of the Intrusion Detection System (IDS) solutions for protecting web servers from HTTP attacks. WAF is a tool to identify and prevent many types of attacks, such as XSS and SQL-injection. In this paper, deep machine learning algorithms are used for enriching the WAF based on the anomaly detection method. Firstly, we construct attributes from HTTP data, to do so we consider two models namely n-gram and one-hot. Then, according to Auto-Encoder LSTM (AE-LSTM) as an unsupervised deep leaning method, we should extract informative features and then reduce them. Finally, we use ensemble isolation forest to train only normal data for the classifier. We apply the proposed model on CSIC 2010 and ECML/ PKDD 2007 datasets. The results show AE-LSTM has higher performance in terms of accuracy and generalization compared with naïve methods on CSIC dataset; the proposed method also have acceptable detection rate on ECML/PKDD dataset using n-gram
Neural Network World
Traffic flow prediction is one of the most interesting machine learning applications in real-worl... more Traffic flow prediction is one of the most interesting machine learning applications in real-world problems that can help anyone move around. In this study, we proposed a feature extraction structure for multivariate time series using Elman recurrent auto-encoder. We added loopback from the encoder layer of the normal auto-encoder to regard sequence information between successive data. The feedback layer implemented using Elman neural network and GRU cells, then the model is trained by different optimization algorithms. The models are also trained using the Emotional Learning method in which we involve the derivative of the error in the cost function to avoid local minimums and keep the last state of the network. We used the proposed method for classification and prediction problems on traffic data from the California Department of Transportation Performance Measurement System (PeMS). The results show that our structure can successfully extract a compact representation of traffic data useful for reconstructing of original data, classification, and prediction. The results also show that adding the recurrent layer to the feature extractor (auto-encoder) leads to better results in the classification phase in comparison with standard methods that do not use the recurrence during feature extraction.
2013 13th Iranian Conference on Fuzzy Systems (IFSC), 2013
This article presents a comparison of Artificial Neural Network (ANN) and Adaptive Neural Fuzzy I... more This article presents a comparison of Artificial Neural Network (ANN) and Adaptive Neural Fuzzy Inference System (ANFIS) for predicting a real system, gold price. Also, we compared a new hybrid model which is a weighted average of the ANN and ANFIS model. The main objective is to predict the gold price in the Forex market. We used two prediction machine models in ANN, a model which feeds back the network output as input and another model that does not do it. Our results show that the performance error of the former model is more than the latter, and also the performance of ANFIS is better than both models of ANN. To evaluate the methods three performance measurements are used: Root Mean Squared Error (RMSE), percentage error and Mean Tendency Error (MTE) which is proposed in this study. The strength point of our method is the prediction machine model that is one of the most powerful prediction machine models of ANN. At last, a Wavelet denoising algorithm is applied to the data, but ...
Aerial images segmentation is a principal task in many applications of remote sensing such as nat... more Aerial images segmentation is a principal task in many applications of remote sensing such as natural disaster monitoring, residential area detection and etc. This paper presents a new method for aerial images segmentation. The method can distinct urban terrains from non-urban terrains using a supervised learning algorithm. Extracted feature for image description is based on principal components analysis of gradient distribution. The proposed method tested on several aerial images of Google Earth taken by satellite and results show that it can segment these images with high accuracy and very fast speed.
International Journal of Information and Communication Technology Research, 2019
Web Application Firewall (WAF) is known as one of the Intrusion Detection System (IDS) solutions ... more Web Application Firewall (WAF) is known as one of the Intrusion Detection System (IDS) solutions for protecting web servers from HTTP attacks. WAF is a tool to identify and prevent many types of attacks, such as XSS and SQL-injection. In this paper, deep machine learning algorithms are used for enriching the WAF based on the anomaly detection method. Firstly, we construct attributes from HTTP data, to do so we consider two models namely n-gram and one-hot. Then, according to Auto-Encoder LSTM (AE-LSTM) as an unsupervised deep leaning method, we should extract informative features and then reduce them. Finally, we use ensemble isolation forest to train only normal data for the classifier. We apply the proposed model on CSIC 2010 and ECML/ PKDD 2007 datasets. The results show AE-LSTM has higher performance in terms of accuracy and generalization compared with naïve methods on CSIC dataset; the proposed method also have acceptable detection rate on ECML/PKDD dataset using n-gram model.
Pattern Analysis and Applications
In this paper we proposed a new method for pedestrian detection in images and videos. Our method ... more In this paper we proposed a new method for pedestrian detection in images and videos. Our method uses a sliding window to search through images. In order to extract the features, each window is divided into overlapping cells and features are extracted from them. The feature that we extracted to describe each window is based on analysis of gradient distribution of each cell. After gradient distribution of a cell computed, the PCA is applied on it and using a mathematical expression that gauges the attitude of edges we got the feature of the cell. Putting the features of the cells next to each other forms the feature vector of the window. Then, the extracted features are classified using Support Vector Machine (SVM). Finally, the learned SVM model tested on the INRIA pedestrian dataset. The proposed method was compared with Histograms of Oriented Gradient (HOG) approach and the results show that our method has comparable detection accuracy as well as having more robustness when facing with noise.
Neural Network World, 2020
Traffic flow prediction is one of the most interesting machine learning applications in real-worl... more Traffic flow prediction is one of the most interesting machine learning applications in real-world problems that can help anyone move around. In this study, we proposed a feature extraction structure for multivariate time series using Elman recurrent auto-encoder. We added loopback from the encoder layer of the normal auto-encoder to regard sequence information between successive data. The feedback layer implemented using Elman neural network and GRU cells, then the model is trained by different optimization algorithms. The models are also trained using the Emotional Learning method in which we involve the derivative of the error in the cost function to avoid local minimums and keep the last state of the network. We used the proposed method for classification and prediction problems on traffic data from the California Department of Transportation Performance Measurement System (PeMS). The results show that our structure can successfully extract a compact representation of traffic data useful for reconstructing of original data, classification, and prediction. The results also show that adding the recurrent layer to the feature extractor (auto-encoder) leads to better results in the classification phase in comparison with standard methods that do not use the recurrence during feature extraction.
8th Iranian conference on Machine Vision and Image Processing (MVIP2013)
In this paper we proposed a new method for pedestrian detection in images and videos. Our method ... more In this paper we proposed a new method for pedestrian detection in images and videos. Our method uses a sliding window to search through images. In order to extract the features, each window is divided into overlapping cells and features are extracted from them. The feature that we extracted to describe each window is based on analysis of gradient distribution of each cell. After gradient distribution of a cell computed, the PCA is applied on it and using a mathematical expression that gauges the attitude of edges we got the feature of the cell. Putting the features of the cells next to each other forms the feature vector of the window. Then, the extracted features are classified using Support Vector Machine (SVM). Finally, the learned SVM model tested on the INRIA pedestrian dataset. The proposed method was compared with Histograms of Oriented Gradient (HOG) approach and the results show that our method has comparable detection accuracy as well as having more robustness when facing with noise.
13th Iraninan Conference on Fuzzy Systems (IFSC2013), Aug 29, 2013
System (ANFIS) for predicting a real system, gold price. Also, we compared a new hybrid model whi... more System (ANFIS) for predicting a real system, gold price. Also, we compared a new hybrid model which is a weighted average of the ANN and ANFIS model. The main objective is to predict the gold price in the Forex market. We used two prediction machine models in ANN, a model which feeds back the network output as input and another model that does not do it. Our results show that the performance error of the former model is more than the latter, and also the performance of ANFIS is better than both models of ANN. To evaluate the methods three performance measurements are used: Root Mean Squared Error (RMSE), percentage error and Mean Tendency Error (MTE) which is proposed in this study. The strength point of our method is the prediction machine model that is one of the most powerful prediction machine models of ANN. At last, a Wavelet denoising algorithm is applied to the data, but due to the chaotic structure of the gold price, it impairs data and causes to reduce the performance of prediction result.
Aerial images segmentation is a principal task in many applications of remote sensing such as nat... more Aerial images segmentation is a principal task in many applications of remote sensing such as natural disaster monitoring, residential area detection and etc. This paper presents a new method for aerial images segmentation. The method can distinct urban terrains from non-urban terrains using a supervised learning algorithm. Extracted feature for image description is based on principal components analysis of gradient distribution. The proposed method tested on several aerial images of Google Earth taken by satellite and results show that it can segment these images with high accuracy and very fast speed.
In this paper the data envelopment analysis model using expected value approach from both optimis... more In this paper the data envelopment analysis model using expected value approach from both optimistic and pessimistic viewpoint introduced. Then this model solved using Genetic algorithm without linearization and decision making units ranked successfully.
2020 8th Iranian Joint Congress on Fuzzy and intelligent Systems (CFIS), 2020
This paper presents the system identification based on a flexible deep neural network for a seven... more This paper presents the system identification based on a flexible deep neural network for a seven degree of freedom(7DOF), a full-car active suspension system that is multi-input and multi-output. The proposed flexible deep neural network, according to input and output data, obtained three layers of flexible auto-encoder. The flexible name was chosen for the learnable activation function parameter in the activation layers. This view permits every neuron to adjust its activation function and adapt the neuron to boost performance. Here flexible tanh activation function introduced, which causes better performance with the same neurons in the hidden layer. The comparison shows the identification error between flexible deep neural network and classical deep neural network. This adaptation, of course, provides prediction improvement.
Web Application Firewall (WAF) is known as one of the Intrusion Detection System (IDS) solutions ... more Web Application Firewall (WAF) is known as one of the Intrusion Detection System (IDS) solutions for protecting web servers from HTTP attacks. WAF is a tool to identify and prevent many types of attacks, such as XSS and SQL-injection. In this paper, deep machine learning algorithms are used for enriching the WAF based on the anomaly detection method. Firstly, we construct attributes from HTTP data, to do so we consider two models namely n-gram and one-hot. Then, according to Auto-Encoder LSTM (AE-LSTM) as an unsupervised deep leaning method, we should extract informative features and then reduce them. Finally, we use ensemble isolation forest to train only normal data for the classifier. We apply the proposed model on CSIC 2010 and ECML/ PKDD 2007 datasets. The results show AE-LSTM has higher performance in terms of accuracy and generalization compared with naïve methods on CSIC dataset; the proposed method also have acceptable detection rate on ECML/PKDD dataset using n-gram
Neural Network World
Traffic flow prediction is one of the most interesting machine learning applications in real-worl... more Traffic flow prediction is one of the most interesting machine learning applications in real-world problems that can help anyone move around. In this study, we proposed a feature extraction structure for multivariate time series using Elman recurrent auto-encoder. We added loopback from the encoder layer of the normal auto-encoder to regard sequence information between successive data. The feedback layer implemented using Elman neural network and GRU cells, then the model is trained by different optimization algorithms. The models are also trained using the Emotional Learning method in which we involve the derivative of the error in the cost function to avoid local minimums and keep the last state of the network. We used the proposed method for classification and prediction problems on traffic data from the California Department of Transportation Performance Measurement System (PeMS). The results show that our structure can successfully extract a compact representation of traffic data useful for reconstructing of original data, classification, and prediction. The results also show that adding the recurrent layer to the feature extractor (auto-encoder) leads to better results in the classification phase in comparison with standard methods that do not use the recurrence during feature extraction.
2013 13th Iranian Conference on Fuzzy Systems (IFSC), 2013
This article presents a comparison of Artificial Neural Network (ANN) and Adaptive Neural Fuzzy I... more This article presents a comparison of Artificial Neural Network (ANN) and Adaptive Neural Fuzzy Inference System (ANFIS) for predicting a real system, gold price. Also, we compared a new hybrid model which is a weighted average of the ANN and ANFIS model. The main objective is to predict the gold price in the Forex market. We used two prediction machine models in ANN, a model which feeds back the network output as input and another model that does not do it. Our results show that the performance error of the former model is more than the latter, and also the performance of ANFIS is better than both models of ANN. To evaluate the methods three performance measurements are used: Root Mean Squared Error (RMSE), percentage error and Mean Tendency Error (MTE) which is proposed in this study. The strength point of our method is the prediction machine model that is one of the most powerful prediction machine models of ANN. At last, a Wavelet denoising algorithm is applied to the data, but ...
Aerial images segmentation is a principal task in many applications of remote sensing such as nat... more Aerial images segmentation is a principal task in many applications of remote sensing such as natural disaster monitoring, residential area detection and etc. This paper presents a new method for aerial images segmentation. The method can distinct urban terrains from non-urban terrains using a supervised learning algorithm. Extracted feature for image description is based on principal components analysis of gradient distribution. The proposed method tested on several aerial images of Google Earth taken by satellite and results show that it can segment these images with high accuracy and very fast speed.
International Journal of Information and Communication Technology Research, 2019
Web Application Firewall (WAF) is known as one of the Intrusion Detection System (IDS) solutions ... more Web Application Firewall (WAF) is known as one of the Intrusion Detection System (IDS) solutions for protecting web servers from HTTP attacks. WAF is a tool to identify and prevent many types of attacks, such as XSS and SQL-injection. In this paper, deep machine learning algorithms are used for enriching the WAF based on the anomaly detection method. Firstly, we construct attributes from HTTP data, to do so we consider two models namely n-gram and one-hot. Then, according to Auto-Encoder LSTM (AE-LSTM) as an unsupervised deep leaning method, we should extract informative features and then reduce them. Finally, we use ensemble isolation forest to train only normal data for the classifier. We apply the proposed model on CSIC 2010 and ECML/ PKDD 2007 datasets. The results show AE-LSTM has higher performance in terms of accuracy and generalization compared with naïve methods on CSIC dataset; the proposed method also have acceptable detection rate on ECML/PKDD dataset using n-gram model.
Pattern Analysis and Applications
In this paper we proposed a new method for pedestrian detection in images and videos. Our method ... more In this paper we proposed a new method for pedestrian detection in images and videos. Our method uses a sliding window to search through images. In order to extract the features, each window is divided into overlapping cells and features are extracted from them. The feature that we extracted to describe each window is based on analysis of gradient distribution of each cell. After gradient distribution of a cell computed, the PCA is applied on it and using a mathematical expression that gauges the attitude of edges we got the feature of the cell. Putting the features of the cells next to each other forms the feature vector of the window. Then, the extracted features are classified using Support Vector Machine (SVM). Finally, the learned SVM model tested on the INRIA pedestrian dataset. The proposed method was compared with Histograms of Oriented Gradient (HOG) approach and the results show that our method has comparable detection accuracy as well as having more robustness when facing with noise.
Neural Network World, 2020
Traffic flow prediction is one of the most interesting machine learning applications in real-worl... more Traffic flow prediction is one of the most interesting machine learning applications in real-world problems that can help anyone move around. In this study, we proposed a feature extraction structure for multivariate time series using Elman recurrent auto-encoder. We added loopback from the encoder layer of the normal auto-encoder to regard sequence information between successive data. The feedback layer implemented using Elman neural network and GRU cells, then the model is trained by different optimization algorithms. The models are also trained using the Emotional Learning method in which we involve the derivative of the error in the cost function to avoid local minimums and keep the last state of the network. We used the proposed method for classification and prediction problems on traffic data from the California Department of Transportation Performance Measurement System (PeMS). The results show that our structure can successfully extract a compact representation of traffic data useful for reconstructing of original data, classification, and prediction. The results also show that adding the recurrent layer to the feature extractor (auto-encoder) leads to better results in the classification phase in comparison with standard methods that do not use the recurrence during feature extraction.
8th Iranian conference on Machine Vision and Image Processing (MVIP2013)
In this paper we proposed a new method for pedestrian detection in images and videos. Our method ... more In this paper we proposed a new method for pedestrian detection in images and videos. Our method uses a sliding window to search through images. In order to extract the features, each window is divided into overlapping cells and features are extracted from them. The feature that we extracted to describe each window is based on analysis of gradient distribution of each cell. After gradient distribution of a cell computed, the PCA is applied on it and using a mathematical expression that gauges the attitude of edges we got the feature of the cell. Putting the features of the cells next to each other forms the feature vector of the window. Then, the extracted features are classified using Support Vector Machine (SVM). Finally, the learned SVM model tested on the INRIA pedestrian dataset. The proposed method was compared with Histograms of Oriented Gradient (HOG) approach and the results show that our method has comparable detection accuracy as well as having more robustness when facing with noise.
13th Iraninan Conference on Fuzzy Systems (IFSC2013), Aug 29, 2013
System (ANFIS) for predicting a real system, gold price. Also, we compared a new hybrid model whi... more System (ANFIS) for predicting a real system, gold price. Also, we compared a new hybrid model which is a weighted average of the ANN and ANFIS model. The main objective is to predict the gold price in the Forex market. We used two prediction machine models in ANN, a model which feeds back the network output as input and another model that does not do it. Our results show that the performance error of the former model is more than the latter, and also the performance of ANFIS is better than both models of ANN. To evaluate the methods three performance measurements are used: Root Mean Squared Error (RMSE), percentage error and Mean Tendency Error (MTE) which is proposed in this study. The strength point of our method is the prediction machine model that is one of the most powerful prediction machine models of ANN. At last, a Wavelet denoising algorithm is applied to the data, but due to the chaotic structure of the gold price, it impairs data and causes to reduce the performance of prediction result.
Aerial images segmentation is a principal task in many applications of remote sensing such as nat... more Aerial images segmentation is a principal task in many applications of remote sensing such as natural disaster monitoring, residential area detection and etc. This paper presents a new method for aerial images segmentation. The method can distinct urban terrains from non-urban terrains using a supervised learning algorithm. Extracted feature for image description is based on principal components analysis of gradient distribution. The proposed method tested on several aerial images of Google Earth taken by satellite and results show that it can segment these images with high accuracy and very fast speed.
In this paper the data envelopment analysis model using expected value approach from both optimis... more In this paper the data envelopment analysis model using expected value approach from both optimistic and pessimistic viewpoint introduced. Then this model solved using Genetic algorithm without linearization and decision making units ranked successfully.