Ste

ANNALS OF THE UNIVERSITY OF CRAIOVA

Series: AUTOMATION, COMPUTERS, ELECTRONICS and MECHATRONICS


    ISSN: 1841-0626 CNCSIS code 11 category B+ internationally indexed, starting with 2010, (Copernicus, Inspec)
    Semiannual publication
    Occasionally, conferences dedicated special issues may be published

    Year 2017 Volume 14 (41) no. 1
  1. Two Lessons on Recurrent Neural Networks. Basic Features and Architectures (pp. 1 - 7)

    Daniela Danciu*
    *Department of Automation and Electronics, University of Craiova, Romania
    (e-mail: ddanciu at automation.ucv.ro)


    Abstract: The main idea of this survey split into two lectures is motivated by the intensive and extensive development of the Recurrent Neural Networks (RNNs) research branch of the Artificial Intelligence (AI) domain. Due to their cyclic interconnections, RNNs are Neural Networks (NNs) which involve dynamics. More specific, RNNs can have very rich spatial and temporal behaviors which include fixed-point multiple equilibria, oscillations (self-oscillations, but also forced oscillations), time-delays, synchronization and even chaotic behaviors. For these reasons RNNs can be used to model complex cognitive functions such as associative memories, but also decision making, classification, sorting as well as formalized problem solving tasks. The first lecture presents RNNs with a special focus on their main features, the artificial neuron and the most used architectures. The second lecture will further discuss the RNNs from the points of view of the qualitative behavior (considering the local properties but also the global behavior) and their main applications.
    Keywords: Recurrent Neural Networks, Hopfield networks, Cohen-Grossberg networks, KWTA neural networks, Cellular Neural Networks, Multiple equilibria.

  2. Overview of Deep Learning in Medical Imaging (pp. 8 - 18)

    Popa(Ciurezu) Didi Liliana*, Faiq Baji*, Popa Radu Teodoru*
    *Department of Computers and Information Technology, University of Craiova, Romania
    (e-mail: lilianaciurezu at yahoo.com, faiq.baji2015 at gmail.com, tpopa at software.ucv.ro)


    Abstract: Deep learning have gained lately popularity by achieving very good results for recognizing objects such as cars, plants, coffee cups in images. Big companies like Facebook, Google, Amazon - are already using these methods to identify faces, recognize voice commands and even enable self-driving cars. Deep learning is based on classical neural networks and represents a method of machine learning and has evolved over the years to become a research field on its own. Deep neural networks are based on different models: Stacked Auto Encoder ,Deep Belief Networks, Deep Boltzmann Machine ,Convolutional Neural Networks, Recurrent Neural Networks. Most deep learning researchers are not programming neural networks directly but, they are using software libraries like: TensorFlow, Caffe2, Theano, Torch, etc. Deep learning is a central method for developing new applications in medical sector. Medical sector has access to vast quantities of patient data and images can be fed in the deep learning neural networks algorithms to learn from. In medical image analysis many types of deep architectures have been applied .In the field of Convolutional Networks there are several architectures. The most common are: LeNet, AlexNet, GoogLeNet, ZfNet, VggNet, ResNet. Today, deep learning networks can execute a lot of tasks in medical field, especially medical imaging. These network can solve problems like: Classification, Regression and Segmentation but they need a lot of data to train deep models and also need powerful hardware to train the deep networks. In this paper are discussed briefly the latest methods for medical imaging currently in research: Blood vessel detection in ultrasound, Classification of skin cancer close to dermatologist level with deep neural networks, Deep CNNs for Diabetic Retinopathy Detection, Deep Learning for large-scale drug screening, Deep Learning Commercial Applications. In conclusion, deep learning has a great potential impact in changing world.
    Keywords: Deep learning, convolutional neural networks, medical imaging, machine learning, medical classification

  3. Initial Results on the Effectiveness of a Skill-Based Approach to Human Resource Allocation (pp. 19 - 24)

    Ionuț Murărețu*, Sorin Ilie*, Mihaela Ilie*
    *Department of Computers and Information Technology, University of Craiova, Romania (e-mail: imuraretu at gmail.com, silie at software.ucv.ro, ela.pirvu at gmail.com)


    Abstract: This paper introduces a skill based approach to human resource allocation. For this purpose a mathematical model is introduced modelling tasks and employees as vectors of skill. Then we present 5 strategies of allocating tasks to employees modeled as fitness functions. These functions are then compared in a simulated environment. The conclusion of the experiment is that allocating tasks to suboptimal employees can speed up the project delivery time at least fivefold. The novelty of this approach is mainly that we opted for a heuristic model of recommendation focused resource allocation that can be applied in continuous task flows.
    Keywords: Mathematical model, Human resource management, Enterprise resource planning, Optimal matching

  4. Technology based on FPGA circuits and simultaneous processing of signals with great dynamic over time (pp. 25 - 30)

    Ion Marian Popescu*, Bogdan Popa*, Răzvan Prejbeanu*
    *Department of Automation and Electronics, University of Craiova, Romania
    (e-mail: pmarian at automation.ucv.ro, bogdan at automation.ucv.ro, razvan.prejbeanu at yahoo.com)


    Abstract: Nowadays according to the paradigm of the Industry 4.0, the IoT (Internet of Things) concept has moved very fast to the IIot (Industrial Internet of Things). This approach claims that the real time processing using a big data can cause a lot of problems on the communication level. This process fills the communication line and the processor with a high number of bytes. All these types of cases are in the field of industrial processes and must be done very fast. A good improvement can be done with the complex processing on the information source and also with the transfer just for the computed data. That means, obviously a low capacity of bytes. In this context, the innovation introduced by National Instruments Company, around 7 years ago with the NI PAC (Programmable Automation Controller) platform suppose an insertion of a FPGA circuit command between the architecture of the embedded processor system and the main data acquisition system. This improvement is useful for the big data processing systems where it has obtained parallel processing possibility for big capacity in fast time processing. In fact, this system is equivalent with a software program but at the hardware level. Also, it is proposed to develop the application and the implementation of the FPGA circuit in a graphical environment such as Labview FPGA Module from the National Instruments Company. This technology will be used to help the Vonrep Company from Targu Jiu, Romania and to monitor a series of processes within this company.
    Keywords: FPGA, Vibration, Real-time, Parallel processing

  5. DC Motor Speed Control Using a Discrete PID Control Law (pp. 31 - 36)

    Teodor-Constantin C. Nichițelea*, Maria-Geanina Gh. Unguritu*
    *Department of Automation and Electronics, University of Craiova, Romania
    (e-mail: teodornichitelea at outlook.com, geany_unguritu at yahoo.com)


    Abstract: This paper presents the speed control of the DC motors of an intelligent racing car by using a discrete-time Proportional-Integral-Derivative controller and tuning methods. Using wireless communication, data is sent from the intelligent car to a MATLAB interface which acquires the received data. The DC motors are therefore modeled using experimental identification.
    Keywords: Discrete-time controller, DC motor control, experimental system identification, PID controller