Search results for: Large scale VLSI neural net

Authors: Zhengsheng Wang, Xiangyong Ji, Yong Du

Abstract:

The projection methods, usually viewed as the methods for computing eigenvalues, can also be used to estimate pseudospectra. This paper proposes a kind of projection methods for computing the pseudospectra of large scale matrices, including orthogonalization projection method and oblique projection method respectively. This possibility may be of practical importance in applications involving large scale highly nonnormal matrices. Numerical algorithms are given and some numerical experiments illustrate the efficiency of the new algorithms.

Keywords: Pseudospectra, eigenvalue, projection method, Arnoldi, IOM(q)

Procedia APA BibTeX Chicago EndNote Harvard JSON MLA RIS XML ISO 690 PDF Downloads 1296

Authors: Myungsook Klassen

Abstract:

Training neural networks to capture an intrinsic property of a large volume of high dimensional data is a difficult task, as the training process is computationally expensive. Input attributes should be carefully selected to keep the dimensionality of input vectors relatively small. Technical indexes commonly used for stock market prediction using neural networks are investigated to determine its effectiveness as inputs. The feed forward neural network of Levenberg-Marquardt algorithm is applied to perform one step ahead forecasting of NASDAQ and Dow stock prices.

Keywords: Stock Market Prediction, Neural Networks, Levenberg-Marquadt Algorithm, Technical Indexes

Procedia APA BibTeX Chicago EndNote Harvard JSON MLA RIS XML ISO 690 PDF Downloads 1921

Authors: M. Tarafdar Haque, A. Taheri

Abstract:

Application of neural networks in execution of programmed pulse width modulation (PPWM) of a voltage source inverter (VSI) is studied in this paper. Using the proposed method it is possible to cancel out the desired harmonics in output of VSI in addition to control the magnitude of fundamental harmonic, contineously. By checking the non-trained values and a performance index, the most appropriate neural network is proposed. It is shown that neural networks may solve the custom difficulties of practical utilization of PPWM such as large size of memory, complex digital circuits and controlling the magnitude of output voltage in a discrete manner.

Keywords: Neural Network, Inverter, PPWM.

Procedia APA BibTeX Chicago EndNote Harvard JSON MLA RIS XML ISO 690 PDF Downloads 1661

Authors: L. Bennaceur Farah, I. R. Farah, R. Bennaceur, Z. Belhadj, M. R. Boussema

Abstract:

The overall objective of this paper is to retrieve soil surfaces parameters namely, roughness and soil moisture related to the dielectric constant by inverting the radar backscattered signal from natural soil surfaces. Because the classical description of roughness using statistical parameters like the correlation length doesn't lead to satisfactory results to predict radar backscattering, we used a multi-scale roughness description using the wavelet transform and the Mallat algorithm. In this description, the surface is considered as a superposition of a finite number of one-dimensional Gaussian processes each having a spatial scale. A second step in this study consisted in adapting a direct model simulating radar backscattering namely the small perturbation model to this multi-scale surface description. We investigated the impact of this description on radar backscattering through a sensitivity analysis of backscattering coefficient to the multi-scale roughness parameters. To perform the inversion of the small perturbation multi-scale scattering model (MLS SPM) we used a multi-layer neural network architecture trained by backpropagation learning rule. The inversion leads to satisfactory results with a relative uncertainty of 8%.

Keywords: Remote sensing, rough surfaces, inverse problems, SAR, radar scattering, Neural networks and Fractals.

Procedia APA BibTeX Chicago EndNote Harvard JSON MLA RIS XML ISO 690 PDF Downloads 1567

Authors: Mohamed A. El-Kady, Badr M. Alshammari

Abstract:

In a recent major industry-supported research and development study, a novel framework was developed and applied for assessment of reliability and quality performance levels in reallife power systems with practical large-scale sizes. The new assessment methodology is based on three metaphors (dimensions) representing the relationship between available generation capacities and required demand levels. The paper shares the results of the successfully completed stud and describes the implementation of the new methodology on practical zones in the Saudi electricity system.

Keywords: Power systems; large-scale analysis, reliability; performance assessment, linear programming.

Procedia APA BibTeX Chicago EndNote Harvard JSON MLA RIS XML ISO 690 PDF Downloads 1821

Authors: Cherif Haddad, Faouzi Ben Charrada

Abstract:

Large-scale systems such as Grids offer infrastructures for both data distribution and parallel processing. The use of Grid infrastructures is a more recent issue that is already impacting the Distributed Database Management System industry. In DBMS, distributed query processing has emerged as a fundamental technique for ensuring high performance in distributed databases. Database placement is particularly important in large-scale systems because it reduces communication costs and improves resource usage. In this paper, we propose a dynamic database placement policy that depends on query patterns and Grid sites capabilities. We evaluate the performance of the proposed database placement policy using simulations. The obtained results show that dynamic database placement can significantly improve the performance of distributed query processing.

Keywords: Large-scale systems, Grid environment, Distributed Databases, Distributed query processing, Database placement

Procedia APA BibTeX Chicago EndNote Harvard JSON MLA RIS XML ISO 690 PDF Downloads 1472

Authors: Hazem M. El-Bakry, Qiangfu Zhao

Abstract:

Recently, fast neural networks for object/face detection were presented in [1-3]. The speed up factor of these networks relies on performing cross correlation in the frequency domain between the input image and the weights of the hidden layer. But, these equations given in [1-3] for conventional and fast neural networks are not valid for many reasons presented here. In this paper, correct equations for cross correlation in the spatial and frequency domains are presented. Furthermore, correct formulas for the number of computation steps required by conventional and fast neural networks given in [1-3] are introduced. A new formula for the speed up ratio is established. Also, corrections for the equations of fast multi scale object/face detection are given. Moreover, commutative cross correlation is achieved. Simulation results show that sub-image detection based on cross correlation in the frequency domain is faster than classical neural networks.

Keywords: Conventional Neural Networks, Fast Neural Networks, Cross Correlation in the Frequency Domain.

Procedia APA BibTeX Chicago EndNote Harvard JSON MLA RIS XML ISO 690 PDF Downloads 2431

Authors: Mountassar Maamoun, Mehdi Neggazi, Abdelhamid Meraghni, Daoud Berkani

Abstract:

This paper presents a VLSI design approach of a highspeed and real-time 2-D Discrete Wavelet Transform computing. The proposed architecture, based on new and fast convolution approach, reduces the hardware complexity in addition to reduce the critical path to the multiplier delay. Furthermore, an advanced twodimensional (2-D) discrete wavelet transform (DWT) implementation, with an efficient memory area, is designed to produce one output in every clock cycle. As a result, a very highspeed is attained. The system is verified, using JPEG2000 coefficients filters, on Xilinx Virtex-II Field Programmable Gate Array (FPGA) device without accessing any external memory. The resulting computing rate is up to 270 M samples/s and the (9,7) 2-D wavelet filter uses only 18 kb of memory (16 kb of first-in-first-out memory) with 256×256 image size. In this way, the developed design requests reduced memory and provide very high-speed processing as well as high PSNR quality.

Keywords: Discrete Wavelet Transform (DWT), Fast Convolution, FPGA, VLSI.

Procedia APA BibTeX Chicago EndNote Harvard JSON MLA RIS XML ISO 690 PDF Downloads 1940

Authors: Anwar Jarndal

Abstract:

In this paper, a genetic-neural-network (GNN) based large-signal model for GaN HEMTs is presented along with its parameters extraction procedure. The model is easy to construct and implement in CAD software and requires only DC and S-parameter measurements. An improved decomposition technique is used to model self-heating effect. Two GNN models are constructed to simulate isothermal drain current and power dissipation, respectively. The two model are then composed to simulate the drain current. The modeling procedure was applied to a packaged GaN-on-Si HEMT and the developed model is validated by comparing its large-signal simulation with measured data. A very good agreement between the simulation and measurement is obtained.

Keywords: GaN HEMT, computer-aided design & modeling, neural networks, genetic optimization.

Procedia APA BibTeX Chicago EndNote Harvard JSON MLA RIS XML ISO 690 PDF Downloads 1621

Authors: Nikhil, Bestamin Özkaya, Ari Visa, Chiu-Yue Lin, Jaakko A. Puhakka, Olli Yli-Harja

Abstract:

The performance of a sucrose-based H2 production in a completely stirred tank reactor (CSTR) was modeled by neural network back-propagation (BP) algorithm. The H2 production was monitored over a period of 450 days at 35±1 ºC. The proposed model predicts H2 production rates based on hydraulic retention time (HRT), recycle ratio, sucrose concentration and degradation, biomass concentrations, pH, alkalinity, oxidation-reduction potential (ORP), acids and alcohols concentrations. Artificial neural networks (ANNs) have an ability to capture non-linear information very efficiently. In this study, a predictive controller was proposed for management and operation of large scale H2-fermenting systems. The relevant control strategies can be activated by this method. BP based ANNs modeling results was very successful and an excellent match was obtained between the measured and the predicted rates. The efficient H2 production and system control can be provided by predictive control method combined with the robust BP based ANN modeling tool.

Keywords: Back-propagation, biohydrogen, bioprocessmodeling, neural networks.

Procedia APA BibTeX Chicago EndNote Harvard JSON MLA RIS XML ISO 690 PDF Downloads 1738

Authors: Vahid Ayatollahi Tafti, Abolfazl Gandomi

Abstract:

A mobile Ad-hoc network consists of wireless nodes communicating without the need for a centralized administration. A user can move anytime in an ad hoc scenario and, as a result, such a network needs to have routing protocols which can adopt dynamically changing topology. To accomplish this, a number of ad hoc routing protocols have been proposed and implemented, which include DSR, OLSR and AODV. This paper presents a study on the QoS parameters for MANET application traffics in large-scale scenarios with 50 and 120 nodes. The application traffics analyzed in this study is File Transfer Protocol (FTP). In large scale networks (120 nodes) OLSR shows better performance and in smaller scale networks (50 nodes)AODV shows less packet drop rate and OLSR shows better throughput.

Keywords: aodv, dsr, manet , olsr , qos.

Procedia APA BibTeX Chicago EndNote Harvard JSON MLA RIS XML ISO 690 PDF Downloads 2138

Authors: Han-hua Lu, Shun-yi Zhang, Yong Zheng, Ya-shi Wang, Li-juan Min

Abstract:

From the perspective of system of systems (SoS) and emergent behaviors, this paper describes large scale application software systems, and proposes framework methods to further depict systems- functional and non-functional characteristics. Besides, this paper also specifically discusses some functional frameworks. In the end, the framework-s applications in system disintegrations, system architecture and stable intermediate forms are additionally dealt with in this in building, deployment and maintenance of large scale software applications.

Keywords: application software system, framework methods, system of systems, emergent behaviors

Procedia APA BibTeX Chicago EndNote Harvard JSON MLA RIS XML ISO 690 PDF Downloads 1364

Authors: Sang-Hun Lee, Bum-Soo Kim, Yang-Sae Moon, Jinho Kim

Abstract:

In this paper we present a photo mosaic smartphone application in client-server based large-scale image databases. Photo mosaic is not a new concept, but there are very few smartphone applications especially for a huge number of images in the client-server environment. To support large-scale image databases, we first propose an overall framework working as a client-server model. We then present a concept of image-PAA features to efficiently handle a huge number of images and discuss its lower bounding property. We also present a best-match algorithm that exploits the lower bounding property of image-PAA. We finally implement an efficient Android-based application and demonstrate its feasibility.

Keywords: smartphone applications; photo mosaic; similarity search; data mining; large-scale image databases.

Procedia APA BibTeX Chicago EndNote Harvard JSON MLA RIS XML ISO 690 PDF Downloads 1651

Authors: T. Hacib, M. R. Mekideche, N. Ferkha

Abstract:

This paper presents an approach which is based on the use of supervised feed forward neural network, namely multilayer perceptron (MLP) neural network and finite element method (FEM) to solve the inverse problem of parameters identification. The approach is used to identify unknown parameters of ferromagnetic materials. The methodology used in this study consists in the simulation of a large number of parameters in a material under test, using the finite element method (FEM). Both variations in relative magnetic permeability and electrical conductivity of the material under test are considered. Then, the obtained results are used to generate a set of vectors for the training of MLP neural network. Finally, the obtained neural network is used to evaluate a group of new materials, simulated by the FEM, but not belonging to the original dataset. Noisy data, added to the probe measurements is used to enhance the robustness of the method. The reached results demonstrate the efficiency of the proposed approach, and encourage future works on this subject.

Keywords: Inverse problem, MLP neural network, parametersidentification, FEM.

Procedia APA BibTeX Chicago EndNote Harvard JSON MLA RIS XML ISO 690 PDF Downloads 1733

Authors: Kexin Jia, Youxin Lu

Abstract:

To make the modulation classification system more suitable for signals in a wide range of signal to noise rate (SNR), a novel method of designing combined classifier based on fuzzy neural network (FNN) is presented in this paper. The method employs fuzzy neural network classifiers and interclass distance (ICD) to improve recognition reliability. Experimental results show that the proposed combined classifier has high recognition rate with large variation range of SNR (success rates are over 99.9% when SNR is not lower than 5dB).

Keywords: Modulation classification, combined classifier, fuzzy neural network, interclass distance.

Procedia APA BibTeX Chicago EndNote Harvard JSON MLA RIS XML ISO 690 PDF Downloads 1200

Authors: Chiou-Yng Lee, Wen-Yo Lee, Chieh-Tsai Wu, Cheng-Chen Yang

Abstract:

Shifted polynomial basis (SPB) is a variation of polynomial basis representation. SPB has potential for efficient bit level and digi -level implementations of multiplication over binary extension fields with subquadratic space complexity. For efficient implementation of pairing computation with large finite fields, this paper presents a new SPB multiplication algorithm based on Karatsuba schemes, and used that to derive a novel scalable multiplier architecture. Analytical results show that the proposed multiplier provides a trade-off between space and time complexities. Our proposed multiplier is modular, regular, and suitable for very large scale integration (VLSI) implementations. It involves less area complexity compared to the multipliers based on traditional decomposition methods. It is therefore, more suitable for efficient hardware implementation of pairing based cryptography and elliptic curve cryptography (ECC) in constraint driven applications.

Keywords: Digit-serial systolic multiplier, elliptic curve cryptography (ECC), Karatsuba algorithm (KA), shifted polynomial basis (SPB), pairing computation.

Procedia APA BibTeX Chicago EndNote Harvard JSON MLA RIS XML ISO 690 PDF Downloads 2036

Authors: Hafiz Muhammad Obaid, Umais Tayyab, Shabbir Majeed Ch.

Abstract:

In this paper, a CMOS differential operational transresistance amplifier (OTRA) is presented. The amplifier is designed and implemented in a standard umc90-nm CMOS technology. The differential OTRA provides wider bandwidth at high gain. It also shows much better rise and fall time and exhibits a very good input current dynamic range of 50 to 50 μA. The OTRA can be used in many analog VLSI applications. The presented amplifier has high gain bandwidth product of 617.6 THz Ω. The total power dissipation of the presented amplifier is also very low and it is 0.21 mW.

Keywords: CMOS, differential, operational transresistance amplifier, OTRA, 90 nm, VLSI.

Procedia APA BibTeX Chicago EndNote Harvard JSON MLA RIS XML ISO 690 PDF Downloads 1113

Authors: Yitao Lei, Xingxiang Zhai, Burra Venkata Durga Kumar

Abstract:

As the quantity and complexity of computing in large-scale software systems increase, distributed system computing becomes increasingly important. The distributed system realizes high-performance computing by collaboration between different computing resources. If there are no efficient resource scheduling resources, the abuse of distributed computing may cause resource waste and high costs. However, resource scheduling is usually an NP-hard problem, so we cannot find a general solution. However, some optimization algorithms exist like genetic algorithm, ant colony optimization, etc. The large scale of distributed systems makes this traditional optimization algorithm challenging to work with. Heuristic and machine learning algorithms are usually applied in this situation to ease the computing load. As a result, we do a review of traditional resource scheduling optimization algorithms and try to introduce a deep reinforcement learning method that utilizes the perceptual ability of neural networks and the decision-making ability of reinforcement learning. Using the machine learning method, we try to find important factors that influence the performance of distributed system computing and help the distributed system do an efficient computing resource scheduling. This paper surveys the application of deep reinforcement learning on distributed system computing resource scheduling. The research proposes a deep reinforcement learning method that uses a recurrent neural network to optimize the resource scheduling. The paper concludes the challenges and improvement directions for Deep Reinforcement Learning-based resource scheduling algorithms.

Keywords: Resource scheduling, deep reinforcement learning, distributed system, artificial intelligence.

Procedia APA BibTeX Chicago EndNote Harvard JSON MLA RIS XML ISO 690 PDF Downloads 435

Authors: Susmita Sahoo, Madhumanti Datta, Rajib Kar

Abstract:

This work proposes an accurate crosstalk noise estimation method in the presence of multiple RLC lines for the use in design automation tools. This method correctly models the loading effects of non switching aggressors and aggressor tree branches using resistive shielding effect and realistic exponential input waveforms. Noise peak and width expressions have been derived. The results obtained are at good agreement with SPICE results. Results show that average error for noise peak is 4.7% and for the width is 6.15% while allowing a very fast analysis.

Keywords: Crosstalk, distributed RLC segments, On-Chip interconnect, output response, VLSI, noise peak, noise width.

Procedia APA BibTeX Chicago EndNote Harvard JSON MLA RIS XML ISO 690 PDF Downloads 1617

Authors: Lihui Ma, Kunlun Xin, Suiqing Liu

Abstract:

Modern managements of water distribution system (WDS) need water quality models that are able to accurately predict the dynamics of water quality variations within the distribution system environment. Before water quality models can be applied to solve system problems, they should be calibrated. Although former researchers use GA solver to calibrate relative parameters, it is difficult to apply on the large-scale or medium-scale real system for long computational time. In this paper a new method is designed which combines both macro and detailed model to optimize the water quality parameters. This new combinational algorithm uses radial basis function (RBF) metamodeling as a surrogate to be optimized for the purpose of decreasing the times of time-consuming water quality simulation and can realize rapidly the calibration of pipe wall reaction coefficients of chlorine model of large-scaled WDS. After two cases study this method is testified to be more efficient and promising, and deserve to generalize in the future.

Keywords: Metamodeling, model calibration, radial basisfunction, water distribution system, water quality model.

Procedia APA BibTeX Chicago EndNote Harvard JSON MLA RIS XML ISO 690 PDF Downloads 1985

Authors: V. H. Mankar, T. S. Das, S. K. Sarkar

Abstract:

In this paper, we have proposed a novel blind watermarking architecture towards its hardware implementation in VLSI. In order to facilitate this hardware realization, cellular automata (CA) concept is introduced. The CA has been already accepted as an attractive structure for VLSI implementation because of its modularity, parallelism, high performance and reliability. The hardware realizable multiresolution spread spectrum watermarking techniques are very few in numbers in spite of their best ever resiliency against signal impairments. This is because of the computational cost and complexity associated with their different filter banks and lifting techniques. The concept of cellular automata theory in order to form a new transform domain technique i.e. Cellular Automata Transform (CAT) have been incorporated. Since CA provides spreading sequences having very low cross-correlation properties, the CA based pseudorandom sequence generator is considered in the present work. Considering the watermarking technique as a digital communication process, an error control coding (ECC) must be incorporated in the data hiding schemes. Besides the hardware implementation of entire CA based data hiding technique, the individual blocks of the algorithm using CA provide the best result than that of some other methods irrespective of the hardware and software technique. The Cellular Automata Transform, CA based PN sequence generator, and CA ECC are the requisite blocks that are developed not only to meet the reliable hardware requirements but also for the basic spread spectrum watermarking features. The proposed algorithm shows statistical invisibility and resiliency against various common signal-processing operations. This algorithmic design utilizes the existing allocated bandwidth in the data transmission channel in a more efficient manner.

Keywords: Cellular automata, watermarking, error control coding, PN sequence, VLSI.

Procedia APA BibTeX Chicago EndNote Harvard JSON MLA RIS XML ISO 690 PDF Downloads 2044

Authors: A. Ben Yaghlane, M. N. Azaiez, M. Mrad

Abstract:

We investigate the large scale of networks in the context of network survivability under attack. We use appropriate techniques to evaluate and the attacker-based- and the defenderbased- network survivability. The attacker is unaware of the operated links by the defender. Each attacked link has some pre-specified probability to be disconnected. The defender choice is so that to maximize the chance of successfully sending the flow to the destination node. The attacker however will select the cut-set with the highest chance to be disabled in order to partition the network. Moreover, we extend the problem to the case of selecting the best p paths to operate by the defender and the best k cut-sets to target by the attacker, for arbitrary integers p,k>1. We investigate some variations of the problem and suggest polynomial-time solutions.

Keywords: Defense/attack strategies, large scale, networks, partitioning a network.

Procedia APA BibTeX Chicago EndNote Harvard JSON MLA RIS XML ISO 690 PDF Downloads 1455

Authors: A. Muthuramalingam, S. Himavathi, E. Srinivasan

Abstract:

.Hardware realization of a Neural Network (NN), to a large extent depends on the efficient implementation of a single neuron. FPGA-based reconfigurable computing architectures are suitable for hardware implementation of neural networks. FPGA realization of ANNs with a large number of neurons is still a challenging task. This paper discusses the issues involved in implementation of a multi-input neuron with linear/nonlinear excitation functions using FPGA. Implementation method with resource/speed tradeoff is proposed to handle signed decimal numbers. The VHDL coding developed is tested using Xilinx XC V50hq240 Chip. To improve the speed of operation a lookup table method is used. The problems involved in using a lookup table (LUT) for a nonlinear function is discussed. The percentage saving in resource and the improvement in speed with an LUT for a neuron is reported. An attempt is also made to derive a generalized formula for a multi-input neuron that facilitates to estimate approximately the total resource requirement and speed achievable for a given multilayer neural network. This facilitates the designer to choose the FPGA capacity for a given application. Using the proposed method of implementation a neural network based application, namely, a Space vector modulator for a vector-controlled drive is presented

Keywords: FPGA implementation, multi-input neuron, neural network, nn based space vector modulator.

Procedia APA BibTeX Chicago EndNote Harvard JSON MLA RIS XML ISO 690 PDF Downloads 4379

Authors: Doshi N. A., Dhobale S. B., Kakade S. R.

Abstract:

As chip manufacturing technology is suddenly on the threshold of major evaluation, which shrinks chip in size and performance, LFSR (Linear Feedback Shift Register) is implemented in layout level which develops the low power consumption chip, using recent CMOS, sub-micrometer layout tools. Thus LFSR counter can be a new trend setter in cryptography and is also beneficial as compared to GRAY & BINARY counter and variety of other applications. This paper compares 3 architectures in terms of the hardware implementation, CMOS layout and power consumption, using Microwind CMOS layout tool. Thus it provides solution to a low power architecture implementation of LFSR in CMOS VLSI.

Keywords: Chip technology, Layout level, LFSR, Pass transistor

Procedia APA BibTeX Chicago EndNote Harvard JSON MLA RIS XML ISO 690 PDF Downloads 4487

Authors: M. Yoneda

Abstract:

In this paper, a lateral walking design force per person is proposed and compared with Imperial College test results. Numerical simulations considering the proposed walking design force which is incorporated into the neural-oscillator model are carried out placing much emphasis on the synchronization (the lock-in phenomenon) for a pedestrian bridge model with the span length of 50 m. Numerical analyses are also conducted for an existing pedestrian suspension bridge. As compared with full scale measurements for this suspension bridge, it is confirmed that the analytical method based on the neural-oscillator model might be one of the useful ways to explain the synchronization (the lock-in phenomenon) of pedestrians being on the bridge.

Keywords: Pedestrian bridge, human-induced lateral vibration, neural-oscillator, full scale measurement, dynamic response analysis.

Procedia APA BibTeX Chicago EndNote Harvard JSON MLA RIS XML ISO 690 PDF Downloads 760

Authors: I. Zamani, M. H. Zarif

Abstract:

In this paper a method for designing of nonlinear controller for a fuzzy model of Double Inverted Pendulum is proposed. This system can be considered as a fuzzy large-scale system that includes offset terms and disturbance in each subsystem. Offset terms are deterministic and disturbances are satisfied a matching condition that is mentioned in the paper. Based on Lyapunov theorem, a nonlinear controller is designed for this fuzzy system (as a model reference base) which is simple in computation and guarantees stability. This idea can be used for other fuzzy large- scale systems that include more subsystems Finally, the results are shown.

Keywords: Controller, Fuzzy Double Inverted Pendulums, Fuzzy Large-Scale Systems, Lyapunov Stability.

Procedia APA BibTeX Chicago EndNote Harvard JSON MLA RIS XML ISO 690 PDF Downloads 2493

Authors: Alexander Goncharenko, Andrey Denisov, Sergey Alyamkin, Evgeny Terentev

Abstract:

The neural network quantization is highly desired procedure to perform before running neural networks on mobile devices. Quantization without fine-tuning leads to accuracy drop of the model, whereas commonly used training with quantization is done on the full set of the labeled data and therefore is both time- and resource-consuming. Real life applications require simplification and acceleration of quantization procedure that will maintain accuracy of full-precision neural network, especially for modern mobile neural network architectures like Mobilenet-v1, MobileNet-v2 and MNAS. Here we present a method to significantly optimize training with quantization procedure by introducing the trained scale factors for discretization thresholds that are separate for each filter. Using the proposed technique, we quantize the modern mobile architectures of neural networks with the set of train data of only ∼ 10% of the total ImageNet 2012 sample. Such reduction of train dataset size and small number of trainable parameters allow to fine-tune the network for several hours while maintaining the high accuracy of quantized model (accuracy drop was less than 0.5%). Ready-for-use models and code are available in the GitHub repository.

Keywords: Distillation, machine learning, neural networks, quantization.

Procedia APA BibTeX Chicago EndNote Harvard JSON MLA RIS XML ISO 690 PDF Downloads 702

Authors: Yavar Safaei Mehrabani, Mehdi Hosseinzadeh

Abstract:

As embedded and portable systems were emerged power consumption of circuits had been major challenge. On the other hand latency as determines frequency of circuits is also vital task. Therefore, trade off between both of them will be desirable. Modulo 2n+1 adders are important part of the residue number system (RNS) based arithmetic units with the interesting moduli set (2n-1,2n, 2n+1). In this manuscript we have introduced novel binary representation to the design of modulo 2n+1 adder. VLSI realization of proposed architecture under 180 nm full static CMOS technology reveals its superiority in terms of area, power consumption and power-delay product (PDP) against several peer existing structures.

Keywords: Computer arithmetic, modulo 2n+1 adders, Residue Number System (RNS), VLSI.

Procedia APA BibTeX Chicago EndNote Harvard JSON MLA RIS XML ISO 690 PDF Downloads 1773

Authors: Saeed H. Nasiri, Rahim Faez, Bita Davoodi, Maryam Farrokhi

Abstract:

Bode stability analysis based on transmission line modeling (TLM) for single wall carbon nanotube (SWCNT) interconnects used in 3D-VLSI circuits is investigated for the first time. In this analysis, the dependence of the degree of relative stability for SWCNT interconnects on the geometry of each tube has been acquired. It is shown that, increasing the length and diameter of each tube, SWCNT interconnects become more stable.

Keywords: Bode stability criterion, Interconnects, Interlayer via, Single wall carbon nanotubes, Transmission line method, Time domain analysis

Procedia APA BibTeX Chicago EndNote Harvard JSON MLA RIS XML ISO 690 PDF Downloads 1801

Authors: Amitabh Wahi, Sundaramurthy S.

Abstract:

Natural outdoor scene classification is active and promising research area around the globe. In this study, the classification is carried out in two phases. In the first phase, the features are extracted from the images by wavelet decomposition method and stored in a database as feature vectors. In the second phase, the neural classifiers such as back-propagation neural network (BPNN) and resilient back-propagation neural network (RPNN) are employed for the classification of scenes. Four hundred color images are considered from MIT database of two classes as forest and street. A comparative study has been carried out on the performance of the two neural classifiers BPNN and RPNN on the increasing number of test samples. RPNN showed better classification results compared to BPNN on the large test samples.

Keywords: BPNN, Classification, Feature extraction, RPNN, Wavelet.

Procedia APA BibTeX Chicago EndNote Harvard JSON MLA RIS XML ISO 690 PDF Downloads 1921