Journal
Scientific and technical journal of information technologies, mechanics and optics
UDK004.932.2
Issue:5 (93)
Download PDF0 Kbyte
IMPROVEMENT OF RECOGNITION QUALITY IN DEEP LEARNING NETWORKS BY SIMULATED ANNEALING METHOD
Annotation
The subject of this research is deep learning methods, in which automatic construction of feature transforms is taken place in tasks of pattern recognition. Multilayer autoencoders have been taken as the considered type of deep learning networks. Autoencoders perform nonlinear feature transform with logistic regression as an upper classification layer. In order to verify the hypothesis of possibility to improve recognition rate by global optimization of parameters for deep learning networks, which are traditionally trained layer-by-layer by gradient descent, a new method has been designed and implemented. The method applies simulated annealing for tuning connection weights of autoencoders while regression layer is simultaneously trained by stochastic gradient descent. Experiments held by means of standard MNIST handwritten digit database have shown the decrease of recognition error rate from 1.1 to 1.5 times in case of the modified method comparing to the traditional method, which is based on local optimization. Thus, overfitting effect doesn’t appear and the possibility to improve learning rate is confirmed in deep learning networks by global optimization methods (in terms of increasing recognition probability). Research results can be applied for improving the probability of pattern recognition in the fields, which require automatic construction of nonlinear feature transforms, in particular, in the image recognition. Keywords: pattern recognition, deep learning, autoencoder, logistic regression, simulated annealing.
Keywords
Постоянный URL
Articles in current issue
- ACCOUNTING OF MANY-PARTICLE INTERACTIONS IN MOLECULAR J-AGGREGATES AND NONLINEAR OPTICAL EFFECTS IN THESE SYSTEMS
- SERS OF BACTERIORHODOPSIN WITH OUT-DIFFUSED SILVER NANOISLANDS
- SOLID BODY ABLATION UNDER EXPOSURE TO ULTRA SHORT LASER PULSES: STUDY BY MOLECULAR DYNAMICS METHODS
- FUNDAMENTAL MATRIX OF LINEAR CONTINUOUS SYSTEM IN THE PROBLEM OF ESTIMATING ITS TRANSPORT DELAY
- THERMAL AND ELECTRIC FIELDS AT SPARK PLASMA SINTERING OF THERMOELECTRIC MATERIALS
- INFLUENCE OF QUARTZ CERAMICS SINGLE-STAGE PROCESSING BY GEL-FORMING WATER SOLUTIONS ON ITS STRENGTH
- INVESTIGATION OF SORPTION CHARACTERISTICS OF POLYMERIC MINERAL-FILLED COMPOSITES FOR MEDICINE
- CRYSTALLIZATION KINETICS OF POLYMERIC NANOCOMPOSITES BASED ON POLYAMIDE 12 MODIFIED BY Cr2O3 NANOPARTICLES
- INORGANIC PHOSPHORS IN GLASS BASED ON LEAD SILICATE GLASSES
- EXPERIMENTAL STUDY OF FIRMWARE FOR INPUT AND EXTRACTION OF USER’S VOICE SIGNAL IN VOICE AUTHENTICATION SYSTEMS
- SYSTEMS FOR SUPPORT OF COLLABORATIVE STUDIES IN THE COLLA ENVIRONMENT BASED ON THE EVENT BUSH METHOD
- BILINGUAL MULTIMODAL SYSTEM FOR TEXT-TO-AUDIOVISUAL SPEECH AND SIGN LANGUAGE SYNTHESIS
- POLICE OFFICE MODEL IMPROVEMENT FOR SECURITY OF SWARM ROBOTIC SYSTEMS
- RELATIONAL THEORY APPLICATION FOR OPTIMAL DESIGN OF INTEGRATED CIRCUITS
- INTERVALS OPTIMIZATION OF SYSTEMS INFORMATION SECURITY INSPECTION
- ALGORITHM FOR SEMANTIC TEXT ANALYSIS BY MEANS OF BASIC SEMANTIC TEMPLATES WITH DELETION
- ARCHITECTURE OF WEB BASED COMPUTER-AIDED MANUFACTURING SYSTEM
- SIMULATION OF PULSED BREAKDOWN IN HELIUM BY ADAPTIVE METHODS
- ON ALGORITHMS CREATION FOR STRAPDOWN STABILIZED GYROCOMPASS OPERATION BASED ON ELECTRICALLY SUSPENDED GYROSCOPE
- MODELING AND EXPERIMENTAL STUDY OF A FIBER OPTIC HYDROPHONE SENSING ELEMENT
- NEW APPROACHES TO EFFICIENCY OF MASSIVE ONLINE COURSE
- INFORMATION INFRASTRUCTURE OF THE EDUCATIONAL ENVIRONMENT WITH VIRTUAL MACHINE TECHNOLOGY
- TWO-LAYER PHASE COMPENSATING INTERFERENCE SYSTEMS
- A PROTOTYPE OF BARENTSNET PROFESSIONAL SOCIAL NETWORK FOR INFORMATION SUPPORT OF DEVELOPMENT MANAGEMENT FOR BARENTS EURO-ARCTIC REGION