• Journal of the Institute of Electronics Engineers of Korea CI
• /
• v.41 no.6
• /
• pp.61-67
• /
• 2004

We propose a new algorithm to generate additional training patterns using the brain-style information processing algorithm, that is, supervised and unsupervised learning models. This will be useful in the case that we do not have enough number of training patterns because of limitation such as time consuming, economic problem, and so on. We adopt the independent component analysis as an unsupervised model for generating exempalr patterns and multilayer perceptions as supervised models for verifying usefulness of the generated patterns. After statistical analysis of the proposed pattern generation algorithm, we verify successful operations of our algorithm through simulation of handwritten digit recognition with various numbers of training patterns.

• The Journal of Korean Institute of Communications and Information Sciences
• /
• v.34 no.2A
• /
• pp.85-90
• /
• 2009

In this paper, as an alternative to constant modulus algorithm based on MSE, maximization of the probability that equalizer output power is equal to the constant modulus of the transmitted symbols is introduced. The proposed algorithm using the gradient ascent method to the maximum probability criterion has superior convergence and steady-state MSE performance, and the error samples of the proposed algorithm exhibit more concentrated density functions in blind equalization environments. Simulation results indicate that the proposed training has a potential advantage versus MSE training for the constant modulus approach to blind equalization.

• The Journal of Korea Robotics Society
• /
• v.16 no.3
• /
• pp.276-282
• /
• 2021

The system for recognizing text in natural scenes has been applied in various industries. However, due to the change in brightness that occurs in nature such as light reflection and shadow, the text recognition performance significantly decreases. To solve this problem, we propose an adversarial shadow generation and training algorithm that is robust to shadow changes. The adversarial shadow generation and training algorithm divides the entire image into a total of 9 grids, and adjusts the brightness with 4 trainable parameters for each grid. Finally, training is conducted in a adversarial relationship between the text recognition model and the shaded image generator. As the training progresses, more and more difficult shaded grid combinations occur. When training with this curriculum-learning attitude, we not only showed a performance improvement of more than 3% in the ICDAR2015 public benchmark dataset, but also confirmed that the performance improved when applied to our's android application text recognition dataset.

• Journal of Biomedical Engineering Research
• /
• v.31 no.5
• /
• pp.365-374
• /
• 2010

This paper is about the personal identification algorithm using an ECG that has been studied by a few researchers recently. Previously published algorithm can be classified as two methods. One is the method that analyzes ECG features and the other is the morphological analysis of ECG. The main characteristic of proposed algorithm uses together two methods. The algorithm consists of training and testing procedures. In training procedure, the features of all recognition objects' ECG were extracted and the PCA was performed for morphological analysis of ECG. In testing procedure, 6 candidate ECG's were chosen by morphological analysis and then the analysis of features among candidate ECG's was performed for final recognition. We choose 18 ECG files from MIT-BIH Normal Sinus Rhythm Database for estimating algorithm performance. The algorithm extracts 100 heartbeats from each ECG file, and use 40 heartbeats for training and 60 heartbeats for testing. The proposed algorithm shows clearly superior performance in all ECG data, amounting to 90.96% heartbeat recognition rate and 100% ECG recognition rate.

• 제어로봇시스템학회:학술대회논문집
• /
• 2003.10a
• /
• pp.1305-1308
• /
• 2003

A new technique is presented to construct a predictive model of plasma etch process. This was accomplished by combining a backpropagation neural network (BPNN) and a genetic algorithm (GA). The predictive model constructed in this way is referred to as a GA-BPNN. The GA played a role of controlling training factors simultaneously. The training factors to be optimized are the hidden neuron, training tolerance, initial weight magnitude, and two gradients of bipolar sigmoid and linear functions. Each etch response was optimized separately. The proposed scheme was evaluated with a set of experimental plasma etch data. The etch process was characterized by a $2^3$ full factorial experiment. The etch responses modeled are aluminum (A1) etch rate, silica profile angle, A1 selectivity, and dc bias. Additional test data were prepared to evaluate model appropriateness. The GA-BPNN was compared to a conventional BPNN. Compared to the BPNN, the GA-BPNN demonstrated an improvement of more than 20% for all etch responses. The improvement was significant in the case of A1 etch rate.

• The Journal of the Acoustical Society of Korea
• /
• v.16 no.1
• /
• pp.104-109
• /
• 1997

This paper derives a new training alorithm for WHMM using the well-known MCE/GPD method with experimental results on the E-set. The derived algorithm generalizes the conventional adaptive training algorithm for WHMM, which means that HMMs of multiple competing classes can be trained at the same time. The recognition results on the E-set have shown about 15% and 12% improvement for training and test data, respectively.

• Journal of Biomedical Engineering Research
• /
• v.15 no.1
• /
• pp.83-88
• /
• 1994

The transform trellis code is an optimal source code as a block size and the constraint length of a shift register go to infinite for stationary Gaussian sources with the squared-error distortion measure. However to implement this code, we have to choose the finite block size and constraint length. Moreover real-world sources are inherently non stationary. To overcome these difficulties, we developed a training algorithm for the transform trellis code. The trained transform trellis code which uses the same rates to each block led to a variation in the resulting distortion from one block to another. To alleviate this non-uniformity in the encoded image, we constructed clusters from the variance of the training data and assigned different rates for each cluster.

• The Journal of Korean Institute of Communications and Information Sciences
• /
• v.22 no.12
• /
• pp.2613-2625
• /
• 1997

Theoretically, a multi-layered feedforward network has been known to be able to approximate a continuous function to an arbitrary degree of accuracy. However, these networks fail to approximate discontinuous functions when they are trained by well-known training algorithms. This paper presents a training algorithm which doesn't work consists of one or more modules, which are trained in a sequential order within subspaces of the input space, and is trained very rapidely once all modules are trained and merged. The experimantal results of applying this method indicates the proposed training algorithm is superior to traditional ones such as baskpagation.

• ETRI Journal
• /
• v.44 no.2
• /
• pp.327-338
• /
• 2022

Classification models for human action recognition require robust features and large training sets for good generalization. However, data augmentation methods are employed for imbalanced training sets to achieve higher accuracy. These samples generated using data augmentation only reflect existing samples within the training set, their feature representations are less diverse and hence, contribute to less precise classification. This paper presents new data augmentation and action representation approaches to grow training sets. The proposed approach is based on two fundamental concepts: virtual video generation for augmentation and representation of the action videos through robust features. Virtual videos are generated from the motion history templates of action videos, which are convolved using a convolutional neural network, to generate deep features. Furthermore, by observing an objective function of the genetic algorithm, the spatiotemporal features of different samples are combined, to generate the representations of the virtual videos and then classified through an extreme learning machine classifier on MuHAVi-Uncut, iXMAS, and IAVID-1 datasets.

• Nuclear Engineering and Technology
• /
• v.55 no.10
• /
• pp.3515-3524
• /
• 2023

A reinforcement learning framework is proposed for the control problem of outlet steam pressure of the once-through steam generator(OTSG) in this paper. The double-layer controller using Proximal Policy Optimization(PPO) algorithm is applied in the control structure of the OTSG. The PPO algorithm can train the neural networks continuously according to the process of interaction with the environment and then the trained controller can realize better control for the OTSG. Meanwhile, reinforcement learning has the characteristic of difficult application in real-world objects, this paper proposes an innovative pretraining method to solve this problem. The difficulty in the application of reinforcement learning lies in training. The optimal strategy of each step is summed up through trial and error, and the training cost is very high. In this paper, the LSTM model is adopted as the training environment for pretraining, which saves training time and improves efficiency. The experimental results show that this method can realize the self-adjustment of control parameters under various working conditions, and the control effect has the advantages of small overshoot, fast stabilization speed, and strong adaptive ability.