• Journal of the Korea Institute of Information and Communication Engineering
• /
• v.25 no.9
• /
• pp.1158-1165
• /
• 2021

This paper proposes a CNN accelerator which is optimized Pooling layer operation incorporated in Multiplication And Accumulation(MAC) to reduce the memory size. For optimizing memory and data path circuit, the quantized 8bit integer weights are used instead of 32bit floating-point weights for pre-training of MNIST data set. To reduce chip area, the proposed CNN model is reduced by a convolutional layer, a 4*4 Max Pooling, and two fully connected layers. And all the operations use specific MAC with approximation adders and multipliers. 94% of internal memory size reduction is achieved by simultaneously performing the convolution and the pooling operation in the proposed architecture. The proposed accelerator chip is designed by using TSMC65nmGP CMOS process. That has about half size of our previous paper, 0.8*0.9 = 0.72mm². The presented CNN accelerator chip achieves 94% accuracy and 77us inference time per an MNIST image.

• Smart Media Journal
• /
• v.7 no.1
• /
• pp.31-36
• /
• 2018

In this paper, we propose automatic screen on / off method of OST-HMD screen using gaze depth estimation technique. The proposed method uses MLP (Multi-layer Perceptron) to learn the user's gaze information and the corresponding distance of the object, and inputs the gaze information to estimate the distance. In the learning phase, eye-related features obtained using a wearable eye-tracker. These features are then entered into the Multi-layer Perceptron (MLP) for learning and model generation. In the inference step, eye - related features obtained from the eye tracker in real time input to the MLP to obtain the estimated depth value. Finally, we use the results of this calculation to determine whether to turn the display of the HMD on or off. A prototype was implemented and experiments were conducted to evaluate the feasibility of the proposed method.

• Journal of the Korea Society of Computer and Information
• /
• v.24 no.11
• /
• pp.51-59
• /
• 2019

Due to the exponential growth of access information on the web, the need for predicting web users' next access has increased. Various models such as markov models, deep neural networks, support vector machines, and fuzzy inference models were proposed to handle web access prediction. For deep learning based on neural network models, training time on large-scale web usage data is very huge. To address this problem, deep neural network models are trained on cluster of computers in parallel. In this paper, we investigated impact of several important spark parameters related to data partitions, shuffling, compression, and locality (basic spark parameters) for training Multi-Layer Perceptron model on Spark standalone cluster. Then based on the investigation, we tuned basic spark parameters for training Multi-Layer Perceptron model and used it for tuning Spark when training Multi-Layer Perceptron model for web access prediction. Through experiments, we showed the accuracy of web access prediction based on our proposed web access prediction model. In addition, we also showed performance improvement in training time based on our spark basic parameters tuning for training Multi-Layer Perceptron model over default spark parameters configuration.

• Journal of the Korean Society for Precision Engineering
• /
• v.31 no.10
• /
• pp.955-965
• /
• 2014

Transport phenomena of reactant and product are directly linked to intrinsic inhomogeneous random configurations of catalyst layer (CL) that consist of ionomer, carbon-supported catalyst (Pt/C), and pores. Hence, electrochemically active surface area (ECSA) of Pt/C is dominated by geometrical morphology of mass transport path. Undoubtedly these ECSAs are key factor of total fuel cell efficiency. In this study, non-deterministic micro-scale CLs were randomly generated by Monte Carlo method and implemented with the percolation process. To ensure valid inference about Pt/C catalyst utilization, 600 samples were chosen as the number of necessary samples with 95% confidence level. Statistic results of 600 samples generated under particular condition (20vol% Pt/C, 30vol% ionomer, 50vol% pore, and 20nm particle diameter) reveal only 18.2%~81.0% of Pt/C can construct ECSAs with mean value of 53.8%. This study indicates that the catalyst utilization in fuel cell CLs cannot be identical notwithstanding the same design condition.

• Proceedings of the KIEE Conference
• /
• 2005.05a
• /
• pp.12-14
• /
• 2005

In this paper, we propose a new architecture of Self-Organizing Fuzzy Polynomial Neural Networks(SOFPNN) by means of genetically optimized fuzzy polynomial neuron(FPN) and discuss its comprehensive design methodology involving mechanisms of genetic optimization, especially genetic algorithms(GAs). The conventional SOFPNNs hinges on an extended Group Method of Data Handling(GMDH) and exploits a fixed fuzzy inference type in each FPN of the SOFPNN as well as considers a fixed number of input nodes located in each layer. The design procedure applied in the construction of each layer of a SOFPNN deals with its structural optimization involving the selection of preferred nodes (or FPNs) with specific local characteristics (such as the number of input variables, the order of the polynomial of the consequent part of fuzzy rules, a collection of the specific subset of input variables, and the number of membership function) and addresses specific aspects of parametric optimization. Therefore, the proposed SOFPNN gives rise to a structurally optimized structure and comes with a substantial level of flexibility in comparison to the one we encounter in conventional SOFPNNs. To evaluate the performance of the genetically optimized SOFPNN, the model is experimented with using two time series data(gas furnace and chaotic time series).

• International Journal of Fuzzy Logic and Intelligent Systems
• /
• v.5 no.4
• /
• pp.327-332
• /
• 2005

We investigatea new fuzzy-neural networks-Hybrid Fuzzy set based polynomial Neural Networks (HFSPNN). These networks consist of genetically optimized multi-layer with two kinds of heterogeneous neurons thatare fuzzy set based polynomial neurons (FSPNs) and polynomial neurons (PNs). We have developed a comprehensive design methodology to determine the optimal structure of networks dynamically. The augmented genetically optimized HFSPNN (namely gHFSPNN) results in a structurally optimized structure and comes with a higher level of flexibility in comparison to the one we encounter in the conventional HFPNN. The GA-based design procedure being applied at each layer of gHFSPNN leads to the selection leads to the selection of preferred nodes (FSPNs or PNs) available within the HFSPNN. In the sequel, the structural optimization is realized via GAs, whereas the ensuing detailed parametric optimization is carried out in the setting of a standard least square method-based learning. The performance of the gHFSPNN is quantified through experimentation where we use a number of modeling benchmarks synthetic and experimental data already experimented with in fuzzy or neurofuzzy modeling.

• Proceedings of the KIEE Conference
• /
• 2001.07d
• /
• pp.2699-2701
• /
• 2001

The study is concerned with an approach to the design of a new category of fuzzy neural networks. The proposed Fuzzy Polynomial Neural Networks(FPNN) with hybrid multi-layer inference architecture is based on fuzzy neural networks(FNN) and polynomial neural networks(PNN) for model identification of complex and nonlinear systems. The one and the other are considered as premise and consequence part of FPNN respectively. We introduce two kinds of FPNN architectures, namely the generic and advanced types depending on the connection points (nodes) of the layer of FNN. Owing to the specific features of two combined architectures, it is possible to consider the nonlinear characteristics of process and to get output performance with superb predictive ability. The availability and feasibility of the FPNN is discussed and illustrated with the aid of two representative numerical examples. The results show that the proposed FPNN can produce the model with higher accuracy and predictive ability than any other method presented previously.

• Proceedings of the KIEE Conference
• /
• 2009.07a
• /
• pp.1703_1704
• /
• 2009

Fuzzy neural ship controllers is used in ship steering control. It can make full use of the advantage of all kinds of intelligent algorithms. This provides an efficient way for this paper. An RBF neural network and GA optimization are employed in a fuzzy neural controller to deal with the nonlinearity, time varying and uncertain factors. Utilizing the designed network to substitute the conventional fuzzy inference, the rule base and membership functions can be auto-adjusted by GA optimization. The parameters of neural network can be decreased by using union-rule configuration in the hidden layer of the network. The ship control quality is effectively improved in case of appending additional sea state disturbance. The performance of controller is evaluated by the system simulation using simulink tools.

• Proceedings of the KIEE Conference
• /
• 2008.07a
• /
• pp.671-672
• /
• 2008

Hybrid intelligent technique is used in ship steering control. It can make full use of the advantage of all kinds of intelligent algorithms. This provides an efficient way for this paper. An RBF neural network and GA optimization are employed in a fuzzy neural controller to deal with the nonlinearity, time varying and uncertain factors. Utilizing the designed network to substitute the conventional fuzzy inference, the rule base and membership functions can be auto-adjusted by GA optimization. The parameters of neural network can be decreased by using union-rule configuration in the hidden layer of the network. The ship control quality is effectively improved in case of appending additional sea state disturbance. The performance of controller is evaluated by the system simulation using Matlab.

• Transactions of the Korean Society of Mechanical Engineers
• /
• v.17 no.7 s.94
• /
• pp.1813-1822
• /
• 1993

An algorithm of fuzzy variable structrue control is proposed to design a closed loop fuel-injection system for the emission control of automotive gasoline engines. Fuzzy control is combined with sliding control at the switching boundary layer to improve the chattering of the stoichiometric air to fuel ratio. Multi-staged fuzzy rules are introduced to improve the adaptiveness of control system for the various operating conditions of engines, and a simplified technique of fuzzy inference is also adopted to improve the computational efficiency based on nonfuzzy micro-processors. The proposed method provides an effective way of engine controller design due to its hybrid structure satisfying the requirements of robustness and stability. The great potential of the fuzzy variable structure control is shown through a hardware-testing with an Intel 80C186 processor for controller and a typical engine-only model on an AD-100 computer.