• Journal of the Korea Institute of Information and Communication Engineering
• /
• v.24 no.3
• /
• pp.355-363
• /
• 2020

In this paper, we propose a deep learning training method for accurately classifying Kuzushiji-MNIST and Kuzushiji-49 datasets for ancient and medieval Japanese characters. We analyze the latest convolutional neural network networks through experiments to select the most suitable network, and then use the networks to select the number of training to classify Kuzushiji-MNIST and Kuzushiji-49 datasets. In addition, the training is conducted with high accuracy by applying learning methods such as Mixup and Random Erase. As a result of the training, the accuracy of the proposed method can be shown to be high by 99.75% for MNIST, 99.07% for Kuzushiji-MNIST, and 97.56% for Kuzushiji-49. Through this deep learning-based technology, it is thought to provide a good research base for various researchers who study East Asian and Western history, literature, and culture.

• Korean Journal of Mathematics
• /
• v.26 no.2
• /
• pp.337-348
• /
• 2018

MNIST dataset is a database containing images of handwritten digits, with each image labeled by an integer from 0 to 9. It is used to benchmark the performance of machine learning algorithms. Neural networks for MNIST are regarded as the starting point of the studying machine learning algorithms. However it is not easy to start the actual programming. In this expository article, we will give a step-by-step instruction to build neural networks for MNIST dataset using MATLAB.

• Journal of IKEEE
• /
• v.22 no.4
• /
• pp.1238-1241
• /
• 2018

In this paper, MNIST classification CNN with zero skipping is implemented. Activation of CNN results in 30% to 40% zero. Since 0 does not affect the MAC operation, skipping 0 through a branch can improve performance. However, at the convolution layer, skipping over a branch causes a performance degradation. Accordingly, in the convolution layer, an operation is skipped by giving a NOP that does not affect the operation. Fully connected layer is skipped through the branch. We have seen performance improvements of about 1.5 times that of existing CNN.

• Journal of the Korea Institute of Information Security & Cryptology
• /
• v.29 no.1
• /
• pp.117-125
• /
• 2019

Deep Learning, which is being used in various fields recently, is being threatened with Adversarial Attack. In this paper, we experimentally verify that the classification accuracy is lowered by adversarial samples generated by malicious attackers in image classification models. We used MNIST dataset and measured the detection accuracy by injecting adversarial samples into the Autoencoder classification model and the CNN (Convolution neural network) classification model, which are created using the Tensorflow library and the Pytorch library. Adversarial samples were generated by transforming MNIST test dataset with JSMA(Jacobian-based Saliency Map Attack) and FGSM(Fast Gradient Sign Method). When injected into the classification model, detection accuracy decreased by at least 21.82% up to 39.08%.

• Journal of IKEEE
• /
• v.25 no.3
• /
• pp.412-419
• /
• 2021

In this paper, we propose a performance improvement of the LwF method using efficient path selection in Continuous Learning Environment. We compare performance and structure with conventional LwF. For comparison, we experiment with performance using MNIST, EMNIST, Fashion MNIST, and CIFAR10 data with different complexity configurations. Experiments show up to 20% improvement in accuracy for each task, which mitigating the Catastrophic Forgetting phenomenon in Continuous Learning environments.

• Journal of information and communication convergence engineering
• /
• v.21 no.2
• /
• pp.139-144
• /
• 2023

Engineers prefer deep neural networks (DNNs) for solving computer vision problems. However, DNNs pose two major problems. First, neural networks require large amounts of well-labeled data for training. Second, the covariate shift problem is common in computer vision problems. Domain adaptation has been proposed to mitigate this problem. Recent work on adversarial-learning-based unsupervised domain adaptation (UDA) has explained transferability and enabled the model to learn robust features. Despite this advantage, current methods do not guarantee the distinguishability of the latent space unless they consider class-aware information of the target domain. Furthermore, source and target examples alone cannot efficiently extract domain-invariant features from the encoded spaces. To alleviate the problems of existing UDA methods, we propose the mixup regularization in adversarial discriminative domain adaptation (ADDA) method. We validated the effectiveness and generality of the proposed method by performing experiments under three adaptation scenarios: MNIST to USPS, SVHN to MNIST, and MNIST to MNIST-M.

• Proceedings of the Korean Society of Computer Information Conference
• /
• 2023.07a
• /
• pp.533-534
• /
• 2023

본 논문에서는 불균형하게 분포된(Non-IID) 데이터를 소유하고 있는 데이터 소유자(클라이언트)들을 가정하고, 데이터 소유자들 간 원본 데이터의 직접적인 이동 없이도 딥러닝 학습이 가능하도록 연합학습을 적용하였다. 실험 환경 구성을 위하여 MNIST 손글씨 데이터 세트를 하나의 숫자만 다량 보유하도록 분할하고 각 클라이언트에게 배포하였다. 연합학습을 적용하여 손글씨 분류 모델을 학습하였을 때 정확도는 85.5%, 중앙집중식 학습모델의 정확도는 90.2%로 연합학습 모델이 중앙집중식 모델 대비 약 95% 수준의 성능을 보여 연합학습 시 성능 하락이 크지 않으며 특수한 상황에서 중앙집중식 학습을 대체할 수 있음을 보였다.

• 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.

• Proceedings of the Korean Institute of Information and Commucation Sciences Conference
• /
• 2019.05a
• /
• pp.439-441
• /
• 2019

Deep learning has become a powerful and robust algorithm in Artificial Intelligence. One of the most impressive forms of Deep learning tools is that of the Convolutional Neural Networks (CNN). CNN is a state-of-the-art solution for object recognition. For instance when we utilize CNN with MNIST handwritten digital dataset, mostly the result is well. Because, in MNIST dataset, all digits are centralized. Unfortunately, the real world is different from our imagination. If digits are shifted from the center, it becomes a big issue for CNN to recognize and provide result like before. To solve that issue, we have created frequency images from spatial images by a Fast Fourier Transform (FFT).

• Proceedings of the Korean Institute of Information and Commucation Sciences Conference
• /
• 2022.05a
• /
• pp.676-677
• /
• 2022

In this paper, we propose a new method based on a deep learning model with an optimized dynamic decay learning rate and improved model structure to achieve fast and accurate classification of fashion clothing images. Experiments are performed using the model proposed in the Fashion-MNIST dataset and compared with methods of CNN, LeNet, LSTM and BiLSTM.