Software & Systems

Articles of journal № 1 at 2021 year.

Order result by: Public date | Title | Authors |

21. The intelligent approach to automation of technological and production processes [№1 за 2021 год]
Authors: S.Yu. Ryabov, Yu.V. Ryabov
Visitors: 3259
The paper considers the approach to production automation, in particular, to the automated design of technological processes. Data processing in existing systems is reduced to a set of rules, and the exe-cuting program in its implementation is like a state machine. Obviously, this approach has its own ceil-ing. It is proposed to represent the production process as something whole, described by an intellectual model. The adopted model of automation of technological and production processes is based on graph theory and graph representation of data and knowledge. The graph is considered as some function of time and computation. It is proposed to use a supergraph as a set of abstract and defined given nodes and abstract and static relations. Thus, every script of the physical reality, every manufacturing situa-tion, considered at any scale, will be modeled as a subgraph of a supergraph. Akka, which is an imple-mentation of an actor computational model, can be an intelligent platform for the implementation of computations. It allows for an intelligent approach to solving the problem of automating production and technological processes. An example of constructing a part of a supergraph for machining a part element is considered by a typical transition-side, including the corresponding tool, processing modes, and a measuring tool. The result of such a system will be a graph with vertices and relations describing the knowledge of techno-logical operations or the state of the production process. The result can be transferred to another sys-tem for execution, saved in the database, or used to analyze the situation.

22. The adaptation of the LSTM neural network model to solve the pattern recognition complex problem [№1 за 2021 год]
Author: V.S. Tormozov
Visitors: 3686
The paper examines the adaptation of the model of artificial neural networks of direct distribution with blocks of long short-term memory (LSTM) for the complex problem of pattern recognition. For artifi-cial neural networks (ANN), the context can be extracted from the input signal vector and from the weight values of the trained network. However, considering the context of a significant volume, the number of neural connections and the complexity of training procedures and network operation in-crease. Instead of receiving context from input values, the context can also be temporarily stored in a special memory buffer, from where it can later be extracted and used as a signal in the ANN's opera-tion. This type of memory is called LSTM. The advantage of networks of this type is that they use memory blocks associated with each neuron of the latent layer, which allows context-related data to be stored when forming recognition patterns. There is the method of linear switching of LSTM units depending on the value of the transmitted signal in the paper. A computational experiment was conducted aimed at investigating the effectiveness of the proposed method and the previously developed neural network of direct distribution of a similar structure. Machine learning was performed for each type of ANN on the same sequence of training ex-amples. The test results were compared for: an ANN of direct propagation, a recurring neural network (RNS) of a similar architecture: with the same number of neurons on each layer, and a network of neu-romodulating interaction with one feedback delay. The optimization criterion, in this case, is the error of the neural network on the training sample, consisting of examples not presented in the test. The effi-ciency of solving the classification problem is evaluated according to two criteria: learning error on the training sample and testing error on the testing sample.

23. Adaptive block-term tensor decomposition in visual question answering systems [№1 за 2021 год]
Authors: M.N. Favorskaya, V.V. Andreev
Visitors: 3221
The paper proposes a method for dimensionality reduction of the internal data representation in deep neural networks used to implement visual question answering systems. Methods of tensor decomposi-tion used to solve this problem in visual question answering systems are reviewed. The problem of these systems is to answer an arbitrary text question about the provided image or video sequence. A technical feature of these systems is the need to combine a visual signal (image or video sequence) with input data in text form. Differences in the features of the input data make it rea-sonable to use different architectures of deep neural networks: most often, a convolutional neural net-work for image processing and a recurrent neural network for text processing. When combining data, the number of model parameters explodes enough so that the problem of finding the most optimal methods for reducing the number of parameters is relevant, even when using modern equipment and considering the predicted growth of computational capabilities. Besides the technical limitations, it should also be noted that an increase in the number of parameters can reduce the model's ability to extract meaningful features from the training set, and increases the likelihood of fitting parameters to insignificant features in the data and "noise". The method of adaptive tensor decomposition proposed in the paper allows, based on training data, optimizing the number of parameters for the block tensor decomposition used for bilinear data fusion. The system was tested and the results were compared with some other visual question-answer systems, in which tensor decomposition methods are used for dimensionality reduction.

24. iLabit OmViSys: A photorealistic simulator based on the omnidirectional camera and structured light [№1 за 2021 год]
Author: Kholodilin, I.Yu.
Visitors: 2992
According to recent advances in neural network learning, which are supported by the demand for large training data, virtual learning has recently attracted a lot of attention from the computer vision commu-nity. Today, there are many available virtual simulation environments, but most of them are based on a standard camera and are limited to measure sensors that are on the mobile robot. To facilitate data collection in systems that were not previously integrated into existing virtual envi-ronments, this paper presents a photorealistic simulator "iLabit OmViSys", which includes an Omnidi-rectional camera, and a structured light source. An Omnidirectional camera and structured light have their own distinctive advantages compared to other computer vision systems. The Omnidirectional camera provides a wide viewing angle with a single shot. In addition, the laser light source is easy to detect and extract its information from this image for further processing. Developed using Unity, the iLabit OmViSys simulator also integrates mobile robots, elements of the internal environment, allows you to generate synthetic photorealistic datasets, and supports communi-cation with third-party programs based on the Transmission Control Protocol (TCP). iLabit OmViSys includes three primary screens that allow one to generate data for internal camera calibration, carried out experiments, and take measurements. A distinctive feature of the simulator is also its versatility, in terms of support for such operating systems as Windows, macOS, and Linux.

25. Application of high-level synthesis technology and hardware accelerators on FPGA in protein identifications [№1 за 2021 год]
Authors: G.K. Shmelev, M.A. Likhachev , Arzhaev V.I.
Visitors: 3184
The paper considers the use of high-level synthesis technology using hardware accelerators based on FPGA in the identifying proteins problem. Currently, there are a significant number of hardware solutions with high performance and band-width designed to solve various applications. One such solution is hardware-based computation accel-erators based on field-programmable gate array (FPGA), which have a number of advantages over ac-celerators built on both graphics processing unit (GPU) and application-specific integrated circuit (ASIC). However, there is a certain complexity in the wide application of such devices, which consists in the laboriousness and specificity of the traditional way of developing applications using specialized programming languages for this type of accelerator. Using high-level synthesis technology using one of the popular programming languages opens up new horizons in the wide use of such accelerators. This paper describes one embodiment of a computational hardware and software platform using a hardware accelerator on a FPGA. Special attention is paid to considering the major steps of developing the architecture of applications deployed on hardware and the methodology for developing a high-performance computing core of hardware-accelerated software functions. The results of improving the computational performance of the de novo peptide sequence sequencing software application and the effectiveness of the used hardware platform and the chosen development path in comparison with the original software application are demonstrated.

◄ ← Preview | 1 | 2 | 3