• ETRI Journal
• /
• v.46 no.4
• /
• pp.660-670
• /
• 2024

The performance of face recognition (FR) has reached a plateau for public benchmark datasets, such as labeled faces in the wild (LFW), celebrities in frontal-profile in the wild (CFP-FP), and the first manually collected, in-the-wild age database (AgeDB), owing to the rapid advances in convolutional neural networks (CNNs). However, the effects of faces under various fine-grained conditions on FR models have not been investigated, owing to the absence of relevant datasets. This paper analyzes their effects under different conditions and loss functions using K-FACE, a recently introduced FR dataset with fine-grained conditions. We propose a novel loss function called MixFace, which combines classification and metric losses. The superiority of MixFace in terms of effectiveness and robustness was experimentally demonstrated using various benchmark datasets.

• KIPS Transactions on Software and Data Engineering
• /
• v.12 no.11
• /
• pp.481-492
• /
• 2023

This study explores how to build a Korean dataset to extract information from text using generative large language models. In modern society, mixed information circulates rapidly, and effectively categorizing and extracting it is crucial to the decision-making process. However, there is still a lack of Korean datasets for training. To overcome this, this study attempts to extract information using text-based zero-shot learning using a generative large language model to build a purposeful Korean dataset. In this study, the language model is instructed to output the desired result through prompt engineering in the form of "system"-"instruction"-"source input"-"output format", and the dataset is built by utilizing the in-context learning characteristics of the language model through input sentences. We validate our approach by comparing the generated dataset with the existing benchmark dataset, and achieve 25.47% higher performance compared to the KLUE-RoBERTa-large model for the relation information extraction task. The results of this study are expected to contribute to AI research by showing the feasibility of extracting knowledge elements from Korean text. Furthermore, this methodology can be utilized for various fields and purposes, and has potential for building various Korean datasets.

• Journal of Computing Science and Engineering
• /
• v.5 no.4
• /
• pp.305-313
• /
• 2011

Due to the expansion of high-speed Internet access, the need for secure and reliable networks has become more critical. The sophistication of network attacks, as well as their severity, has also increased recently. As such, more and more organizations are becoming vulnerable to attack. The aim of this research is to classify network attacks using neural networks (NN), which leads to a higher detection rate and a lower false alarm rate in a shorter time. This paper focuses on two classification types: a single class (normal, or attack), and a multi class (normal, DoS, PRB, R2L, U2R), where the category of attack is also detected by the NN. Extensive analysis is conducted in order to assess the translation of symbolic data, partitioning of the training data and the complexity of the architecture. This paper investigates two engines; the first engine is the back-propagation neural network intrusion detection system (BPNNIDS) and the second engine is the radial basis function neural network intrusion detection system (BPNNIDS). The two engines proposed in this paper are tested against traditional and other machine learning algorithms using a common dataset: the DARPA 98 KDD99 benchmark dataset from International Knowledge Discovery and Data Mining Tools. BPNNIDS shows a superior response compared to the other techniques reported in literature especially in terms of response time, detection rate and false positive rate.

• International Journal of Internet, Broadcasting and Communication
• /
• v.11 no.4
• /
• pp.31-36
• /
• 2019

Unsupervised neural networks have not caught enough attention until Generative Adversarial Network (GAN) was proposed. By using both the generator and discriminator networks, GAN can extract the main characteristic of the original dataset and produce new data with similarlatent statistics. However, researchers understand fully that training GAN is not easy because of its unstable condition. The discriminator usually performs too good when helping the generator to learn statistics of the training datasets. Thus, the generated data is not compelling. Various research have focused on how to improve the stability and classification accuracy of GAN. However, few studies delve into how to improve the training efficiency and to save training time. In this paper, we propose a novel optimizer, named FAST-ADAM, which integrates the Lookahead to ADAM optimizer to train the generator of a semi-supervised generative adversarial network (SSGAN). We experiment to assess the feasibility and performance of our optimizer using Canadian Institute For Advanced Research - 10 (CIFAR-10) benchmark dataset. From the experiment results, we show that FAST-ADAM can help the generator to reach convergence faster than the original ADAM while maintaining comparable training accuracy results.

• 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.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• v.13 no.4
• /
• pp.1795-1811
• /
• 2019

In order to achieve rapid and accurate detection of vehicle objects in complex traffic conditions, we propose a novel vehicle detection method. Firstly, more contextual and small-object vehicle information can be obtained by our Joint Feature Network (JFN). Secondly, our Evolved Region Proposal Network (EPRN) generates initial anchor boxes by adding an improved version of the region proposal network in this network, and at the same time filters out a large number of false vehicle boxes by soft-Non Maximum Suppression (NMS). Then, our Mask Network (MaskN) generates an example that includes the vehicle occlusion, the generator and discriminator can learn from each other in order to further improve the vehicle object detection capability. Finally, these candidate vehicle detection boxes are optimized to obtain the final vehicle detection boxes by the Fine-Tuning Network(FTN). Through the evaluation experiment on the DETRAC benchmark dataset, we find that in terms of mAP, our method exceeds Faster-RCNN by 11.15%, YOLO by 11.88%, and EB by 1.64%. Besides, our algorithm also has achieved top2 comaring with MS-CNN, YOLO-v3, RefineNet, RetinaNet, Faster-rcnn, DSSD and YOLO-v2 of vehicle category in KITTI dataset.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• v.16 no.2
• /
• pp.632-657
• /
• 2022

Detecting web attacks is a major challenge, and it is observed that the use of simple models leads to low sensitivity or high false positive problems. In this study, we aim to develop a robust two-stage deep learning based stacked ensemble web application firewall. Normal and abnormal classification is carried out in the first stage of the proposed WAF model. The classification process of the types of abnormal traffics is postponed to the second stage and carried out using an integrated stacked ensemble model. By this way, clients' requests can be served without time delay, and attack types can be detected with high sensitivity. In addition to the high accuracy of the proposed model, by using the statistical similarity and diversity analyses in the study, high generalization for the ensemble model is achieved. Within the study, a comprehensive, up-to-date, and robust multi-class web anomaly dataset named GAZI-HTTP is created in accordance with the real-world situations. The performance of the proposed WAF model is compared to state-of-the-art deep learning models and previous studies using the benchmark dataset. The proposed two-stage model achieved multi-class detection rates of 97.43% and 94.77% for GAZI-HTTP and ECML-PKDD, respectively.

• Nuclear Engineering and Technology
• /
• v.55 no.2
• /
• pp.603-622
• /
• 2023

Human error (HE) is an important concern in safety-critical systems such as nuclear power plants (NPPs). HE has played a role in many accidents and outage incidents in NPPs. Despite the increased automation in NPPs, HE remains unavoidable. Hence, the need for HE detection is as important as HE prevention efforts. In NPPs, HE is rather rare. Hence, anomaly detection, a widely used machine learning technique for detecting rare anomalous instances, can be repurposed to detect potential HE. In this study, we develop an unsupervised anomaly detection technique based on generative adversarial networks (GANs) to detect anomalies in manually collected surveillance data in NPPs. More specifically, our GAN is trained to detect mismatches between automatically recorded sensor data and manually collected surveillance data, and hence, identify anomalous instances that can be attributed to HE. We test our GAN on both a real-world dataset and an external dataset obtained from a testbed, and we benchmark our results against state-of-the-art unsupervised anomaly detection algorithms, including one-class support vector machine and isolation forest. Our results show that the proposed GAN provides improved anomaly detection performance. Our study is promising for the future development of artificial intelligence based HE detection systems.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• v.9 no.3
• /
• pp.1155-1172
• /
• 2015

In this paper, we propose a novel saliency detection framework based on illumination invariant features to improve the accuracy of the saliency detection under the different light conditions. The proposed algorithm is divided into three steps. First, we extract the illuminant invariant features to reduce the effect of the illumination based on the local sensitive histograms. Second, a preliminary saliency map is obtained in the CIE Lab color space. Last, we use the region growing method to fuse the illuminant invariant features and the preliminary saliency map into a new framework. In addition, we integrate the information of spatial distinctness since the saliency objects are usually compact. The experiments on the benchmark dataset show that the proposed saliency detection framework outperforms the state-of-the-art algorithms in terms of different illuminants in the images.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• v.7 no.4
• /
• pp.819-833
• /
• 2013

Computing global features based on local features using a bio-inspired framework has shown promising performance. However, for some tough applications with large intra-class variances, a single local feature is inadequate to represent all the attributes of the images. To integrate the complementary abilities of multiple local features, in this paper we have extended the efficacy of the bio-inspired framework, HMAX, to adapt heterogeneous features for global feature extraction. Given multiple global features, we propose an approach, designated as parameterized metric learning, for high dimensional feature fusion. The fusion parameters are solved by maximizing the canonical correlation with respect to the parameters. Experimental results show that our method achieves significant improvements over the benchmark bio-inspired framework, HMAX, and other related methods on the Caltech dataset, under varying numbers of training samples and feature elements.