mohammad ghasemzadeh

Artificial neural networks are among the most significant models in machine learning that use numeric inputs. This study presents a new single-layer perceptron model based on categorical inputs. In the proposed model, every quality value in the training dataset receives a trainable weight. Input data is classified by determining the weight vector that corresponds to the categorical values in it. To evaluate the performance of the proposed algorithm, we have used 10 datasets. We have compared the performance of the proposed method to that of other machine learning models, including neural networks, support vector machines, naïve Bayes classifiers, and random forests. According to the results, the proposed model resulted in a 36% reduction in memory usage when compared to baseline models across all datasets. Moreover, it demonstrated a training speed enhancement of 54.5% for datasets that contained more than 1000 samples. The accuracy of the proposed model is also comparable to other machine learning models.

This research is related to the use of deep learning tools and image processing technology in the automatic generation of images from text. Previous researches have used one sentence to produce images. In this research, a memory-based hierarchical model is presented that uses three different descriptions that are presented in the form of sentences to produce and improve the image. The proposed scheme focuses on using more information to produce high-resolution images, using competitive productive networks. Implementing programs related to this field require massive processing resources. Therefore, the proposed method was implemented and tested on a cluster with 25 GPUs using the hardware platform of the University of Copenhagen. The experiments were performed on CUB-200 and ids-ade datasets. The experimental results show that the proposed model can produce higher quality images than the two basic models StackGAN and AttGAN.

Part of speech tagging is an important issue in natural language processing and is the base of many other major subjects in this field. In this article, a new method have been introduced for part of speech tagging using deep neural networks. The purpose of this method is solving problems that common other methods are facing with which are extract deep features from texts and classifying these features. The proposed method is based on that we can find deep features and product optimal output by using small deep neural networks. This method was implemented using Tensorflow's specialized libraries and Keras API in python and was evaluated on coNLL2000 standard dataset. The experimental results show that the proposed method is capable to extract high level features from natural language's words and is able to achieve considerable accuracy for repetitive tags. In addition, this method is able to be used in variant environments and on different devices.

The biological knowledge of different species can be transferred to the conserved sequence regions via genomic sequence alignment. Similarly, through biological network alignment, knowledge of the conserved regions of molecular networks can be transferred to different conserved regions of different species. Therefore, relying on biological network alignment, we can extend the traditional "sequence-based homology" concept to the new concept of "network-based homology". Discovery of networks alignment is especially important because of its applications, such as discovering new drugs, tracking disease progression, or predicting users' behavior on social networks. In this regard, the main challenge is that the problem of finding the alignments in two graphs is NP-hard. In situations like this, we can use the relatively fast meta-heuristic algorithms to find some acceptable approximate solutions. The main contribution of this research consists of conducting a comparison over the network alignment algorithms, based on the respective evaluation criteria, execution time, memory consumption and complexity of the testing networks. The experimental results are obtained from running the relevant algorithms on the well-known BioGRID dataset. Evaluations indicate that among other methods, using genetic algorithm, memetic, particle swarm optimization, simulated annealing and the ant colony, could yield more valuable results. The named methods apply appropriate heuristics to generate and investigate only a very small subset of the whole search space with the highest probability of holding a solution; therefore, they often can find the optimal solution or some acceptable solutions in a relatively short time.

The decision tree is one of the popular methods for learning and reasoning through recursive partitioning of data space. To choose the best attribute in the case on numerical features, partitioning criteria should be calculated for individual values or the value range of each attribute should be divided into two or more intervals using a set of cut points. In partitioning range of attribute, the fuzzy partitioning can be used to reduce the noise sensitivity of data and to increase the stability of decision trees. Since the tree-building algorithms need to keep in main memory the whole training dataset, they have memory restrictions. In this paper, we present an algorithm that builds the fuzzy decision tree on the large dataset. In order to avoid storing the entire training dataset in main memory and overcome the memory limitation, the algorithm builds DTs in an incremental way. In the discretization stage, a fuzzy partition was generated on each continuous attribute based on fuzzy entropy. Then, in order to select the best feature for branches, two criteria, including fuzzy information gain and occurrence matrix are used. Besides, real datasets are used to evaluate the behavior of the algorithm in terms of classification accuracy, decision tree complexity, and execution time as well. The results show that proposed algorithm without a need to store the entire dataset in memory and reduce the complexity of the tree is able to overcome the memory limitation and making balance between accuracy and complexity .

Promoting security and maintaining privacy policies is one of the most important demands in all areas, including the Internet and its specific aspects, such as the Internet of Things (IoT). To achieve this, several solutions have been proposed. In this article, we show how to get valuable results on the Internet using software defined networks (SDN). In this regard, I will first review the concepts of IoT and SDN, and then we'll come up with a way to integrate the IoT and SDN so that we can quickly identify and discontinue threatening attacks and unsecured connections. Meanwhile, we put such an attacker on a blacklist so that they will not be allowed to connect in the future. To evaluate the operation of the proposed method, we need several objects connected to the Internet of objects and an active software network. Theoretical analysis suggests that the impact of this approach on network readiness against attacking the data of devices connected to the Internet of objects and overcoming DDoS attacks is very significant.

Long-term prediction of time series is an important but challenging issue. Today, deep networks, especially Long short Term Memory (LSTM) networks, have been successfully used to predict time series. The LSTM network is capable of maintaining long-term dependencies, but its ability to assign varying degrees of attention to sub window features over multiple time steps is not sufficient. Also, the performance of these networks depends heavily on their hyper-parameters ​​and it is important to adopt an efficient method to ensure optimum values. In this study, to overcome the above challenges, an evolutionary attention-based deep LSTM for predicting multivariate time series is recommended that automatically finds one of the best combinations of LSTM parameter values and sub window features. The proposed algorithm uses a genetic algorithm to properly adjust the deep LSTM network architecture. In order to evaluate the performance of the proposed algorithm, three data sets in the fields of energy and environment have been used. The experimental results show that the proposed algorithm performs better than other basic models.

Both Ti-6Al-4V and 316L stainless steels are widely used as engineering alloys in many modern industries, especially aerospace, gas turbine engines and heat exchangers due to high mechanical properties, high creep, fatigue, and corrosion resistance. Fusion welding of these alloys is not easily possible due to their incomplete solubility in each other. Brazing is one of the best choices for joining dissimilar alloys. Infrared brazing is a novel process characterized with a high heating rate up to 500 °C/min. Accordingly, it is expected that both erosion of the substrate and excessive growth of intermetallic phases in the brazed joint are significantly decreased. This study is focused on the infrared brazing of Ti-6Al-4V and 316L stainless steel to investigate the influence of brazing parameters (brazing alloy, temperature and time) on microstructure and mechanical properties of base metals with CBS 34 (Ag-based) braze filler foils. Brazing was performed in a home-made infrared furnace at temperatures of 750, 780, 800, 850 and 900 °C for 3-5 min. Qualities of the brazed joints were evaluated by microstructure and phase constitution of the bonded joints with light optical microscope (LOM) and scanning electron microscope (SEM). Mechanical properties of the brazed joints were evaluated by hardness test. The optimum brazing parameters were at 850 °C-5 min for CBS 34 filler foils.

This research shows how we can find Balanced Spanning Trees, without the need to determine the relevant parameters, by using an evolutionary multi-objective algorithm. This problem belongs to the complexity class NP-complete; therefore, for big graphs, we cannot use exact algorithms, which often rely on exhaustive search. The existing approximate algorithms are all limited to obtaining the related parameter(s) and considering them separately during their computations; this usually leads to finding solution with lower quality than desired. On the other hand, the existing methods solve the problem of the balanced spanning tree as a single objective; which loss search space information and finding only one solution, while in real-world problems, decision-makers often need several solutions to make a better decision. Overcoming the above-mentioned shortcomings leads to finding a balanced spanning tree with better characteristics; which in turn, yields more benefits in relevant applications, such as communication networks and high-performance computing projects. In this research, in order to overcome the above challenge, the use of evolutionary multi-objective optimization via Non-dominated Sorting Genetic Algorithm - NSGA is recommended. The proposed solution uses two objective functions to simultaneously optimize the weight of the spanning tree and the shortest path. The first objective function attempts to minimize the distance between each vertex and the root of the tree. The second objective function is selected to minimize the weight of the obtained spanning tree. The proposed method was implemented and run in a Python environment by specialized functions, on a seven-core microcomputer. In order to evaluate the performance of the proposed algorithm, a set of random graphs by Erdos and Renyi approaches were used. The proposed algorithm was compared with the state of the art methods in the problem of finding the balanced spanning tree. Experimental results show that the proposed algorithm is often able to find better responses than the other algorithms compared. The experimental results show that the proposed algorithm was always able to find highly ranked Balanced Spanning Trees. Meanwhile, often the optimal solutions, which can only be obtained through exact and very costly algorithms, were found, with teeny computations.

Unmanned Aerial vehicles (UAVs) emerged into a promising research trend applied in several disciplines based on the benefits, including efficient communication, on-time search, and rescue operations, appreciate customer deliveries among more. The current technologies are using fixed base stations (BS) to operate onsite and off-site in the fixed position with its associated problems like poor connectivity. These open gates for the UAVs technology to be used as a mobile alternative to increase accessibility in beam selection with a fifth-generation (5G) connectivity that focuses on increased availability and connectivity. This paper presents a first fast semi-online 3-Dimensional machine learning algorithm suitable for proper beam selection as is emitted from UAVs. Secondly, it presents a detailed step by step approach that is involved in the multi-armed bandit approach in solving UAV solving selection exploration to exploitation dilemmas. The obtained results depicted that a multi-armed bandit problem approach can be applied in optimizing the performance of any mobile networked devices issue based on bandit samples like Thompson sampling, Bayesian algorithm, and ε-Greedy Algorithm. The results further illustrated that the 3-Dimensional algorithm optimizes utilization of technological resources compared to the existing single and the 2-Dimensional algorithms thus close optimal performance on the average period through machine learning of realistic UAV communication situations.

For decades, numbering schemes have been used to track different versions and revisions of files, software packages, documents, and other types of informational objects. The absence of an all-purpose, adaptive, comprehensive and efficient standard makes various problems in different applications of information technology. To help address these issues, this article presents a quadruple and adaptive scheme for version numbering. In this scheme, the version ID is created using four integers. We call these four numbers from left to right "Release Sequence Number", "Generation Number", "Features List Number" and "Corrections List Number." We have defined specific values for the quadrilateral numbers and their meanings. In addition, the criteria for changing the quadruple numbers are specified in different application types. This is an "adaptive" scheme, as it truly has the ability to track, various versions of files, software packages, project output documents, plans, rules, bylaws, style sheets, maps, graphics, administrative and legal documents, and other similar things in different environments without need of restructuring. Also, a new inventive method called "Extended Zero Leading" is introduced which is used to achieve "lexicographical ordering" in names containing the version ID. The scheme can be used to observe changes in the many types of informational objects such as virtual machine memory in the live migration process. Experimental and analytical results indicate the appropriateness and effectiveness of the proposed scheme in meeting the expectations. Although we have used the proposed scheme for research in virtual machine live migration, it can be used as a comprehensive and effective scheme in all administrative, regulatory, legal, academic, engineering, manufacturing, industrial, document production, operational, and software development environments. It can become a universal standard and be applied successfully. The standardization of the proposed scheme and its widespread usage can also be a great help to harmonize everyone's understanding of the version and revision numbering of informational objects.

Various numbering schemes are used to track different versions and revisions of files, software packages, and documents. One major challenge in this regard is the lack of an all-purpose, adaptive, comprehensive and efficient standard. To resolve the challenge, this article presents Quadruple Adaptive Version Numbering Scheme. In the proposed scheme, the version identifier consists of four integers. These four numbers from Left to Right are called: "Release Sequence Number", "Generation Number", "Features List Number", and "Corrections List Number" respectively. In the article, special values are given for the quadruple numbers and their meanings are described. QAVNS is an "Adaptive" scheme; this means that it has the capability to track the different versions and revisions of files, software packages, project output documents, design documents, rules, manuals, style sheets, drawings, graphics, administrative and legal documents, and the other types of "Informational Objects" in different environments, without alterations in its structure. The proposed scheme has the capability to monitor changes in the types of informational objects, such as virtual machine memory, in the live migration process. The experimental and analytical results indicate the desirability and effectiveness of the proposed scheme in satisfying the desired expectations. The proposed scheme can become a common standard and successfully applied in all academic, engineering, administrative, legislative, legal, manufacturing, industrial, operational, software development, documentary, and other environments. The standardization of this scheme and its widespread usage can be a great help in improving everyone's understanding of the numbering of versions & revisions.

Genetic and morphologic similarities between mouse embryonic stem cells (ESCs) and primordial germ cells (PGCs) make it difficult to distinguish differentiation of these two cell types in vitro. Using specific GC markers expressed in low level or even not expressed in ESCs- can help recognize differentiated cells in vitro. We attempted to differentiate the mouse ESCs into Gc-like cells spontaneously in monolayer and EB culture method. In this experimental study, we attempted to differentiate ESCs, Oct4-GFP OG2, into GC-like cells (GCLCs) spontaneously in two different ways, including: i. Spontaneous differentiation of ESCs in monolayer culture as (SP) and ii. Spontaneous differentiation of ESCs using embryoid body (EB) culture method as (EB+SP). During culture, expression level of four GC specific genes (Fkbp6, Mov10l1, Riken and Tex13) and Mvh, Scp3, Stra8, Oct4 were evaluated. In both groups, Mov10l1 was down-regulated (P=0.3), while Tex13 and Riken were up-regulated (P=0.3 and P=0.04, respectively). Fkbp6 and Stra8 were decreased in EB+SP and they were increased in SP group, while no significant difference was determined between them (P=0.1, P=0.07). Additionally, in SP group, gene expression of Mvh and Scp3 were up-regulated and they had significant differences compared to EB+SP group (P=0.00 and P=0.01, respectively). Oct4 was down-regulated in the both groups. Flow-cytometry analysis showed that mean number of Mvh-positive cells in the SP group was significantly greater compared to ESCs, EB+SP and EB7 groups (P=0.00, P=0.01, and P=0.3, respectively). These findings showed that ESCs were differentiated into GCLCs in both group. But spontaneous dif- ferentiation of ESCs into GCLCs in SP group (monolayer culture) compared to EB+SP (EB culture methods) has more ability to express GCs markers.

Systematic studies indicate a growing number of clinical studies that use mesenchymal stem cells (MSCs) for the treatment of cartilage lesions. The current experimental and preclinical study aims to comparatively evaluate the potential of MSCs from a variety of tissues for the treatment of cartilage defect in rabbit’s knee which has not previously been reported. In this experimental study, MSCs isolated from bone marrow (BMMSCs), adipose (AMSCs), and ears (EMSCs) of rabbits and expanded under in vitro culture. The growth rate and differentiation ability of MSCs into chondrocyte and the formation of cartilage pellet were investigated by drawing the growth curve and real-time polymerase chain reaction (RT-PCR), respectively. Then, the critical cartilage defect was created on the articular cartilage (AC) of the rabbit distal femur, and MSCs in collagen carrier were transplanted. The studied groups were as the control (only defect), sham (defect with scaffold), BMMSCs in the scaffold, EMSCs in the scaffold, and EMSCs in the scaffold with cartilage pellets. Histological and the gene expression analysis were performed following the transplantation. Based on our comparative in vitro investigation, AMSCs possessed the highest growth rate, as well as the lowest chondrogenic differentiation potential. In this context, MSCs of the ear showed a significantly higher growth rate and cartilage differentiation potential than those of bone marrow tissue (P<0.05). According to our in vivo assessments, BMMSC- and EMSC-seeded scaffolds efficiently improved the cartilage defect 4 weeks post-transplantation, while no improvement was observed in the group contained the cartilage pellets. It seems that the ear contains MSCs that promote cartilage regeneration as much as the conventional MSCs from the bone marrow. Considering a high proliferation rate and easy harvesting of MSCs of the ear, this finding could be of value for the regenerative medicine.

In this research work, the impact of user’s behavior on search engine results is discussed. It aims to improvement of search results which leads to the higher satisfaction of users. In other words, we are trying to present a personalized search engine for each user, based on his/her activity and search history. We base our hypothesis that the search history of each user in a specific time frame provides precious information to be used in case of offering customizable and more efficient results to the user. In order to evaluate our research project hypothesis, we designed and implemented an experimental search engine as a web platform. This search engine measures the level of user’s satisfaction by bringing putting extra information into account. According to the experimental results, consideration of the user’s behavior history has significant effect on the quality of the search results, leading to more satisfaction of the users.

Medical imaging comprises different imaging modalities and processes to image human body for diagnostic and treatment purposes and، therefore has an important role in the improvement of public health in all population groups. In this paper، we present an intelligent hybrid noise reduction filter which is based on Neuro-Fuzzy systems. It is especially beneficial in medical image noise reduction. First stage we feed the input image into four general noise reduction filters in parallel. These general filters are: mean filter، median filter، weighted median filter and the adaptive median filter. At the second stage we give the output of the above filters as input into a Neuro-Fuzzy system. As expected، the ability of Neuro-Fuzzy systems in encoding human knowledge and using non-deterministic knowledge، allow us to achieve much more noise reduction from the input images. We implement the proposed method and use it for reduction of noise from a set of medical images affected with high noise density. Experimental results show that the idea is considerably effective.

This paper shows how we can make advantage of using genetic programming in selection of suitable features for automatic modulation recognition. Automatic modulation recognition is one of the essential components of modern receivers. In this regard, selection of suitable features may significantly affect the performance of the process. In this research we implemented our model by using appropriate software and hardware platforms. Simulations were conducted with 5db and 10db SNRs. We generated test and training data from real ones recorded in an actual communication system. For performance analysis of the proposed method a set of experiments were conducted considering signals with 2ASK, 4ASK, 2PSK, 4PSK, 2FSK and 4FSK modulations. The results show that the selected features by the suggested model improve the performance of automatic modulation recognition considerably. During our experiments we also reached the optimum values and forms for mutation and crossover ratio, elitism policy, fitness function as well as other parameters for the proposed model.

Quality of service (QoS) is an important issue in the design and management of web service composition. QoS in web services consists of various non-functional factors, such as execution cost, execution time, availability, successful execution rate, and security. In recent years, the number of available web services has proliferated, and then offered the same services increasingly. The same web services are distinguished based on their quality parameters. Also, clients usually demand more value added services rather than those offered by single, isolated web services. Therefore, selecting a composition plan of web services among numerous plans satisfies client requirements and has become a challenging and time-consuming problem. This paper has proposed a new composition plan optimizer with constraints based on genetic algorithm. The proposed method can find the composition plan that satisfies user constraints efficiently. The performance of the method is evaluated in a simulated environment.

Nowadays one of the aspects of learning is using internet and other information technology products. Already some efforts have been accomplished in using Internet services for teaching and learning different science materials، including religious instructions. One of these efforts was concerned with establishment of question answering systems، where specialists could give answers to the users questions. RASEKHOON is a web site which comprises more than eighteen thousand religious questions with their corresponding answers. In practice، users must look for a question which mostly match with what they are concerned about. The aim of this research is to develop a system that takes a question from a user، then it tries to find the most similar questions in the database to the asked question. The searching procedure benefits from an extension of a language model which is based on intelligent information retrieval and natural language processing techniques. It lets users find answers to theirs questions very rapidly.

Binary Decision Diagram (BDD) is a data structure proved to be compact in representation and efficient in manipulation of Boolean formulas. Using Binary decision diagram in network reliability analysis has already been investigated by some researchers. In this paper we show how an exact algorithm for network reliability can be improved and implemented efficiently by using CUDD - Colorado University Decision Diagram.