Computer Science
Loading...
ISSN 1508-2806
e-ISSN: 2300-7036
Issue Date
2020
Volume
Vol. 21
Number
No. 1
Description
Reviewed by: Samuel Sujith, Laouamer Lamri, Adel Khelifi, Marcin Kurdziel, Marcin Kuta, Hamed Yazdapanah, Filip Malawski, Ghada Saad, Tomasz Krzeszowski, Bogdan Kwolek, Renata Slota, Krzysztof Wolk
Journal Volume
Computer Science
Vol. 21 (2020)
Projects
Pages
Articles
Towards a distributed solution to multi-robot task allocation problem with energetic and spatiotemporal constraints
(Wydawnictwa AGH, 2020) Zitouni, Farouq; Harous, Saad; Maamri, Ramdane
This paper tackles the Multi-Robot Task Allocation problem. It consists of two distinct sets: a set of tasks (requiring resources), and a set of robots (offering resources). Then, the tasks are allocated to robots while optimizing a certain objective function subject to some constraints, e.g., allocating the maximum number of tasks, minimizing the distances traveled by the robots, etc. Previous works mainly optimized the temporal and spatial constraints, but no work focused on energetic constraints. Our main contribution is the introduction of energetic constraints on multi-robot task allocation problems. In addition, we propose an allocation method based on parallel distributed guided genetic algorithms and compare it to two state-of-the-art algorithms. The performed simulations and obtained results show the effectiveness and scalability of our solution, even in the case of a large number of robots and tasks. We believe that our contribution is applicable in many contemporary areas of research such as smart cities and related topics.
Compressing sentiment analysis CNN models for efficient hardware processing
(Wydawnictwa AGH, 2020) Wróbel, Krzysztof; Karwatowski, Michał; Wielgosz, Maciej; Pietroń, Marcin; Wiatr, Kazimierz
Convolutional neural networks (CNNs) were created for image classification tasks. Shortly after their creation, they were applied to other domains, including natural language processing (NLP). Nowadays, solutions based on artificial intelligence appear on mobile devices and embedded systems, which places constraints on memory and power consumption, among others. Due to CNN memory and computing requirements, it is necessary to compress them in order to be mapped to the hardware. This paper presents the results of the compression of efficient CNNs for sentiment analysis. The main steps involve pruning and quantization. The process of mapping the compressed network to an FPGA and the results of this implementation are described. The conducted simulations showed that the 5-bit width is enough to ensure no drop in accuracy when compared to the floating-point version of the network. Additionally, the memory footprint was significantly reduced (between 85 and 93% as compared to the original model).
Data censoring with set-membership affine projection algorithm
(Wydawnictwa AGH, 2020) Karamali, Gholamreza; Zardadi, Akram; Moradi, Hamid Reza
In this work, we use the single-threshold and double-threshold set-membership affine projection algorithm to censor non-informative and irrelevant data in big data problems. For this purpose, we employ the probability distribution function of the additive noise in the desired signal and the excess of the meansquared error (EMSE) in steady-state to evaluate the threshold parameter of the single -threshold set-membership affine projection (ST-SM-AP) algorithm intending to obtain the desired update percentage. In addition, we propose the double-threshold set-membership affine projection (DT-SM-AP) algorithm to detect very large errors caused by unrelated data (such as outliers). The DT-SM-AP algorithm is capable of censoring non-informative and unrelated data in big data problems, and it will promote the misalignment and convergence speed of the learning procedure with low computational complexity. The synthetic examples and real-life experiments substantiate the superior performance of the proposed algorithms as compared to traditional algorithms.
Extraction of scores and average from Algerian high-school degree transcripts
(Wydawnictwa AGH, 2020) Kefali, Abderrahmane; Drabsia, Soumia; Sari, Toufik; Chaoui, Mohammed; Ferkous, Chokri
A system for extracting scores and the average from Algerian high school degree transcripts is proposed. The system extracts the scores and average based on the localization of tables gathering this information, it consists of several stages. After preprocessing, the system locates the tables using ruling-line information as well as other text information. Therefore, the adopted localization approach can work even in the absence of certain ruling lines or the erasure and discontinuity of the lines. After this, the localized tables are segmented into columns and the columns into information cells. Finally, cell labeling is done based on prior knowledge of the table structure, allowing us to identify the scores and the average. Experiments have been conducted on a local dataset in order to evaluate the performances of our system and compare it to three public systems at three levels, the obtained results show the effectiveness of our system.
Object pose estimation in monocular image using modified FDCM
(Wydawnictwa AGH, 2020) Dabbour, Abd Alrazzak; Habib, Rabie; Saii, Mariam
In this paper, a new method for multi-object detection and pose estimation in a monocular image is proposed based on the FDCM method. This method can detect an object with a high-speed running time even if the object was under partial occlusion or bad illumination. Additionally, it only requires a single template without any training process. In this paper, a new method (MFDCM) for 3D multi-object pose estimation in a monocular image is proposed, which is based on the FDCM method with major performance improvements in accuracy and running time. These improvements were achieved by using the LSD method instead of a simple edge detector (Canny detector), using an angular Voronoi diagram instead of calculating the 3D distance transform image, a distance transform image, and an integral distance transform image at each orientation. In addition, the search process in the proposed method depends on a line segment-based search instead of the sliding window search in the FDCM. As a result, the proposed method is more robust and much faster than the FDCM method, and the position, scale, and rotation are invariant. In addition, the proposed method was evaluated and compared to different methods (COF, HALCON, LINE2D, and BOLD) using a D-textureless dataset. The comparison results show that the MFDCM has the highest score among all of the tested methods (with a slight advantage from the COF and BLOD methods) while it was a little slower than LINE2D (which was the fasted method among the compared methods). Furthermore, it was at least 14-times faster than the FDCM in the tested scenarios. The results prove that the MFDCM is able to detect and 3D pose estimate of object in a clear or clustered background from a monocular image with a high-speed running time, even if the objects are under partial occlusion, this makes it robust and reliable for real-time applications.