Digital Library[ Search Result ]
CLS Token Additional Embedding Method Using GASF and CNN for Transformer based Time Series Data Classification Tasks
Jaejin Seo, Sangwon Lee, Wonik Choi
http://doi.org/10.5626/JOK.2023.50.7.573
Time series data refer to a sequentially determined data set collected for a certain period of time. They are used for prediction, classification, and outlier detection. Although existing artificial intelligence models in the field of time series are mainly based on the Recurrent Neural Network, recent research trends are changing to transformer based models. Although these transformer based models show good performance for time series data prediction problem, they show relatively insufficient performance for classification tasks. In this paper, we propose an embedding method to add special classification tokens generated using Gramian Angular Summation Field and Convolution Neural Network to utilize time series data as input to transformers and found that we could leverage the pre-trained method to improve performance. To show the efficacy of our method, we conducted extensive experiments with 12 different models using the University of California, Riverside dataset. Experimental results show that our proposed model improved the average accuracy of 85 datasets from 1.4% to up to 21.1%.
R-FLHE: Robust Federated Learning Framework Against Untargeted Model Poisoning Attacks in Hierarchical Edge Computing
http://doi.org/10.5626/JOK.2023.50.1.94
Federated learning is a server-client based distributed learning strategy that collects only trained model to guarantee data privacy and reduce communication costs. Recently, research is being conducted to prepare for the future IoT ecosystem by combining edge computing and federated learning. However, research considering vulnerabilities and threat is insufficient. In this paper, we propose Robust Federated Learning in Hierarchical Edge computing (R-FLHE), a federated learning framework for robust global model from untargeted model poisoning attacks. R-FLHE can aggregate models learned from clients, evaluate them on the edge server, and score them based on the calculated model’s loss. R-FLHE can maintain robustness of the global model by sending only the model of the edge server with the best score to the cloud server. The R-FLHE proposed in this paper shows robustness in maintaining constant performance for each federated learning round, with performance drop of only 0.81% and 1.88% on average even if attacks occur.
Automatic Classification of Pneumonia Based on Ensemble Deep Learning Model Using Intensity Normalization and Multiscale Lung-Focused Patches on Chest X-Ray Images
Yoon Jo Kim, Jinseo An, Helen Hong
http://doi.org/10.5626/JOK.2022.49.9.677
It is difficult to classify normal and pneumonia in pediatric chest X-ray (CXR) images due to irregular intensity values. In addition, deep learning model has a limitation in that it can misclassify CXR by incorrectly focusing on the outer part of the lung. This study proposed an automatic classification of pneumonia based on ensemble deep learning model using three intensity normalizations and multiscale lung-focused patches on CXR images. First, to correct for irregular intensity values in internal lungs, three intensity normalization methods were performed respectively. Second, to focus on internal lungs, regions of interest were extracted by segmenting lung regions. Third, multiscale lung-focused patches were extracted to train the characterization of pneumonia. Finally, ensemble modeling with attention module was performed to improve the classification performance. In the experiment, the method using large patches of CLAHE images showed an accuracy of 92%, which was 5% higher than that of original images. Furthermore, the proposed method using an ensemble of large and middle patches showed the best performance with an accuracy of 93%.
CNN-based Reduced Complexity Decision Confidence Estimation for Imbalanced Web Application Attack Detection
Seungyoung Park, Hansung Kim, Taejoon Jung
http://doi.org/10.5626/JOK.2020.47.9.842
As web application attacks have been rapidly increasing and their types have been diversified, there are limitations on detecting them with the existing schemes. To resolve this problem, the detection techniques using machine learning such as the convolutional neural network (CNN) have been proposed. However, the confidence on the decision error sample in these techniques has been unreliable. To estimate more reliable decision confidence, the Monte-Carlo batch normalization (MCBN) technique combined with the CNN has been proposed. In particular, the CNN performs multiple decisions on a given evaluation sample using multiple mini-batches containing it. Then, its decision confidence estimate is obtained by averaging the multiple decision results. However, it requires too large of a computational load. The reason is that each mini-batch comprises randomly selected (M-1) training samples and only one evaluation sample, when the mini-batch size is M. In this paper, we propose a reduced complexity decision confidence estimation scheme for imbalanced web application attack detection. Specifically, the proposed scheme reduces the computational load by up to M times compared to the MCBN scheme. Also, at the estimation process, the ratio of normal and attack samples in the mini-batch should be maintained the same as that of the training process. To achieve this, we found which class size was small by performing a temporal decision on the evaluation samples. Then, the small class was over-sampled using the training samples to maintain the ratio. Our experimental results showed that the performance improved, and the reliability estimation performance was not significantly degraded compared to the MCBN scheme.
Efficient CNNs with Channel Attention and Group Convolution for Facial Expression Recognition
MyeongOh Lee, Ui Nyoung Yoon, Seunghyun Ko, Geun-Sik Jo
http://doi.org/10.5626/JOK.2019.46.12.1241
Recently, studies using the convolutional neural network have been actively conducted to recognize emotions from facial expressions. In this paper, we propose an efficient convolutional neural network that solves the model complexity problem of the deep convolutional neural network used to recognize the emotions in facial expression. To reduce the complexity of the model, we used group convolution, depth-wise separable convolution to reduce the number of parameters, and the computational cost. We also enhanced the reuse of features and channel information by using Skip Connection for feature connection and Channel Attention. Our method achieved 70.32% and 85.23% accuracy on FER2013, RAF-single datasets with four times fewer parameters (0.39 Million, 0.41 Million) than the existing model.
Effect Scene Detection using Multimodal Deep Learning Models
Jeongseon Lim, Mikyung Han, Hyunjin Yoon
http://doi.org/10.5626/JOK.2018.45.12.1250
A conventional movie can be converted into a 4D movie by identifying effect scenes. In order to automate this process, in this paper, we propose a multimodal deep learning model that detects effect scenes using both visual and audio features of a movie. We have classified effect/non-effect scenes using audio-based Convolutional Recurrent Neural Network (CRNN) model and video-based Long Short-term Memory (LSTM) and Multilayer Perceptron (MLP) model. Also, we have implemented feature-level fusion. In addition, based on our own observation that effects typically occur during non-dialog scenes, we further detected non-dialog scenes using audio-based Convolutional Neural Network (CNN) model. Subsequently, the prediction scores of audio-visual effect scene classification and audio-based non-dialog classification models were combined. Finally, we detected sequences of effect scenes of the entire movie using prediction score of the input window. Experiments using real-world 4D movies demonstrate that the proposed multimodal deep learning model outperforms unimodal models in terms of effect scene detection accuracy.
Web Application Attack Detection Scheme Using Convolutional Neural Networks
Yeongung Seo, Myungjin Kim, Seungyoung Park, Seokwoo Lee
http://doi.org/10.5626/JOK.2018.45.7.744
Because rates of web application attacks are rapidly increasing, web application attack detection schemes using machine learning have recently become of interest. Existing schemes, however, require the selection of a suitable set of features representing the characteristics of expected attacks, and this set of features needs to be adjusted every time a new type of attack is discovered. In this paper, we propose a web application attack detection scheme employing a convolutional neural network (CNN) without the need to select any features in advance. Specifically, the CNN is trained in a supervised manner with images transformed from hexadecimally converted characters in HTTP traffic, without any restriction in the input characters used. Our experimental results show that the proposed scheme improves detection error rate performance by up to 84.4% over existing schemes.
Hybrid Word-Character Neural Network Model for the Improvement of Document Classification
http://doi.org/10.5626/JOK.2017.44.12.1290
Document classification, a task of classifying the category of each document based on text, is one of the fundamental areas for natural language processing. Document classification may be used in various fields such as topic classification and sentiment classification. Neural network models for document classification can be divided into two categories: word-level models and character-level models that treat words and characters as basic units respectively. In this study, we propose a neural network model that combines character-level and word-level models to improve performance of document classification. The proposed model extracts the feature vector of each word by combining information obtained from a word embedding matrix and information encoded by a character-level neural network. Based on feature vectors of words, the model classifies documents with a hierarchical structure wherein recurrent neural networks with attention mechanisms are used for both the word and the sentence levels. Experiments on real life datasets demonstrate effectiveness of our proposed model.
Investigating the Feature Collection for Semantic Segmentation via Single Skip Connection
http://doi.org/10.5626/JOK.2017.44.12.1282
Since the study of deep convolutional neural network became prevalent, one of the important discoveries is that a feature map from a convolutional network can be extracted before going into the fully connected layer and can be used as a saliency map for object detection. Furthermore, the model can use features from each different layer for accurate object detection: the features from different layers can have different properties. As the model goes deeper, it has many latent skip connections and feature maps to elaborate object detection. Although there are many intermediate layers that we can use for semantic segmentation through skip connection, still the characteristics of each skip connection and the best skip connection for this task are uncertain. Therefore, in this study, we exhaustively research skip connections of state-of-the-art deep convolutional networks and investigate the characteristics of the features from each intermediate layer. In addition, this study would suggest how to use a recent deep neural network model for semantic segmentation and it would therefore become a cornerstone for later studies with the state-of-the-art network models.
A Transfer Learning Method for Solving Imbalance Data of Abusive Sentence Classification
http://doi.org/10.5626/JOK.2017.44.12.1275
The supervised learning approach is suitable for classification of insulting sentences, but pre-decided training sentences are necessary. Since a Character-level Convolution Neural Network is robust for each character, so is appropriate for classifying abusive sentences, however, has a drawback that demanding a lot of training sentences. In this paper, we propose transfer learning method that reusing the trained filters in the real classification process after the filters get the characteristics of offensive words by generated abusive/normal pair of sentences. We got higher performances of the classifier by decreasing the effects of data shortage and class imbalance. We executed experiments and evaluations for three datasets and got higher F1-score of character-level CNN classifier when applying transfer learning in all datasets.
Search
Journal of KIISE
- ISSN : 2383-630X(Print)
- ISSN : 2383-6296(Electronic)
- KCI Accredited Journal
Editorial Office
- Tel. +82-2-588-9240
- Fax. +82-2-521-1352
- E-mail. chwoo@kiise.or.kr