• Journal of Korea Multimedia Society
• /
• v.21 no.12
• /
• pp.1459-1466
• /
• 2018

Robots which detect human and recognize action are important factors for human interaction, and many researches have been conducted. Recently, deep learning technology has developed and learning based robot's technology is a major research area. These studies require a database to learn and evaluate for intelligent human perception. In this paper, we propose a multi-modal sensor-based image database condition considering the security task by analyzing the image database to detect the person in the outdoor environment and to recognize the behavior during the running of the robot.

• The Korean Journal of Applied Statistics
• /
• v.32 no.1
• /
• pp.41-68
• /
• 2019

Multi-view learning considers data from various viewpoints as well as attempts to integrate various information from data. Multi-view learning has been studied recently and has showed superior performance to a model learned from only a single view. With the introduction of deep learning techniques to a multi-view learning approach, it has showed good results in various fields such as image, text, voice, and video. In this study, we introduce how multi-view learning methods solve various problems faced in human behavior recognition, medical areas, information retrieval and facial expression recognition. In addition, we review data integration principles of multi-view learning methods by classifying traditional multi-view learning methods into data integration, classifiers integration, and representation integration. Finally, we examine how CNN, RNN, RBM, Autoencoder, and GAN, which are commonly used among various deep learning methods, are applied to multi-view learning algorithms. We categorize CNN and RNN-based learning methods as supervised learning, and RBM, Autoencoder, and GAN-based learning methods as unsupervised learning.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• v.16 no.7
• /
• pp.2407-2424
• /
• 2022

Recently, Transformer has made great progress in video retrieval tasks due to its high representation capability. For the structure of a Transformer, the cascaded self-attention modules are capable of capturing long-distance feature dependencies. However, the local feature details are likely to have deteriorated. In addition, increasing the depth of the structure is likely to produce learning bias in the learned features. In this paper, an improved Transformer structure named TransDCS (Transformer with Dynamic Convolution and Shortcut) is proposed. A Multi-head Conv-Self-Attention module is introduced to model the local dependencies and improve the efficiency of local features extraction. Meanwhile, the augmented shortcuts module based on a dual identity matrix is applied to enhance the conduction of input features, and mitigate the learning bias. The proposed model is tested on MSRVTT, LSMDC and Activity-Net benchmarks, and it surpasses all previous solutions for the video-text retrieval task. For example, on the LSMDC benchmark, a gain of about 2.3% MdR and 6.1% MnR is obtained over recently proposed multimodal-based methods.

• The Journal of Korea Robotics Society
• /
• v.8 no.3
• /
• pp.217-227
• /
• 2013

We introduce a learning system for the sight reading of simple drum sequences. Sight reading is a cognitive-motor skill that requires reading of music symbols and actions of multiple limbs for playing the music. The system provides knowledge of results (KR) pertaining to the learner's performance by color-coding music symbols, and guides the learner by indicating the corresponding action for a given music symbol using additional auditory or vibrotactile cues. To evaluate the effects of KR and guidance cues, three learning methods were experimentally compared: KR only, KR with auditory cues, and KR with vibrotactile cues. The task was to play a random 16-note-long drum sequence displayed on a screen. Thirty university students learned the task using one of the learning methods in a between-subjects design. The experimental results did not show statistically significant differences between the methods in terms of task accuracy and completion time.

• Proceedings of the Korea Information Processing Society Conference
• /
• 2023.11a
• /
• pp.514-517
• /
• 2023

본 연구는 인 컨택스트 러닝 (In-Context Learning)을 오디오-언어 작업에 적용하기 위한 멀티모달 (Multi-Modal) 딥러닝 모델을 다룬다. 해당 모델을 통해 학습 단계에서 오디오와 텍스트의 소통 가능한 형태의 표현 (Representation)을 학습하고 여러가지 오디오-텍스트 작업을 수행할 수 있는 멀티모달 딥러닝 모델을 개발하는 것이 본 연구의 목적이다. 모델은 오디오 인코더와 언어 인코더가 연결된 구조를 가지고 있으며, 언어 모델은 6.7B, 30B 의 파라미터 수를 가진 자동회귀 (Autoregressive) 대형 언어 모델 (Large Language Model)을 사용한다 오디오 인코더는 자기지도학습 (Self-Supervised Learning)을 기반으로 사전학습 된 오디오 특징 추출 모델이다. 언어모델이 상대적으로 대용량이기 언어모델의 파라미터를 고정하고 오디오 인코더의 파라미터만 업데이트하는 프로즌 (Frozen) 방법으로 학습한다. 학습을 위한 과제는 음성인식 (Automatic Speech Recognition)과 요약 (Abstractive Summarization) 이다. 학습을 마친 후 질의응답 (Question Answering) 작업으로 테스트를 진행했다. 그 결과, 정답 문장을 생성하기 위해서는 추가적인 학습이 필요한 것으로 보였으나, 음성인식으로 사전학습 한 모델의 경우 정답과 유사한 키워드를 사용하는 문법적으로 올바른 문장을 생성함을 확인했다.

• Journal of the Korea Society of Computer and Information
• /
• v.27 no.11
• /
• pp.29-38
• /
• 2022

This paper proposes a system for classifying gait types using an ensemble deep learning network for gait data measured by a smart insole equipped with multi-sensors. The gait type classification system consists of a part for normalizing the data measured by the insole, a part for extracting gait features using a deep learning network, and a part for classifying the gait type by inputting the extracted features. Two kinds of gait feature maps were extracted by independently learning networks based on CNNs and LSTMs with different characteristics. The final ensemble network classification results were obtained by combining the classification results. For the seven types of gait for adults in their 20s and 30s: walking, running, fast walking, going up and down stairs, and going up and down hills, multi-sensor data was classified into a proposed ensemble network. As a result, it was confirmed that the classification rate was higher than 90%.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• v.12 no.1
• /
• pp.392-412
• /
• 2018

Due to the semantic gap problem across different modalities, automatically retrieval from multimedia information still faces a main challenge. It is desirable to provide an effective joint model to bridge the gap and organize the relationships between them. In this work, we develop a deep image annotation and classification by fusing multi-modal semantic topics (DAC_mmst) model, which has the capacity for finding visual and non-visual topics by jointly modeling the image and loosely related text for deep image annotation while simultaneously learning and predicting the class label. More specifically, DAC_mmst depends on a non-parametric Bayesian model for estimating the best number of visual topics that can perfectly explain the image. To evaluate the effectiveness of our proposed algorithm, we collect a real-world dataset to conduct various experiments. The experimental results show our proposed DAC_mmst performs favorably in perplexity, image annotation and classification accuracy, comparing to several state-of-the-art methods.

• The Journal of Society for e-Business Studies
• /
• v.27 no.1
• /
• pp.15-28
• /
• 2022

Due to recent expansion of online market such as clothing, utilizing customer review has become a major marketing measure. User review has been used as a tool of analyzing sentiment of customers. Sentiment analysis can be largely classified with machine learning-based and lexicon-based method. Machine learning-based method is a learning classification model referring review and labels. As research of sentiment analysis has been developed, multi-modal models learned by images and video data in reviews has been studied. Characteristics of words in reviews are differentiated depending on products' and customers' categories. In this paper, sentiment is analyzed via considering review data and metadata of products and users. Gated Recurrent Unit (GRU), Long Short-Term Memory (LSTM), Self Attention-based Multi-head Attention models and Bidirectional Encoder Representation from Transformer (BERT) are used in this study. Same Multi-Layer Perceptron (MLP) model is used upon every products information. This paper suggests a multi-modal sentiment analysis model that simultaneously considers user reviews and product meta-information.

• Journal of the Korea Institute of Information and Communication Engineering
• /
• v.23 no.1
• /
• pp.27-32
• /
• 2019

Research that implements the classification process through Deep Learning algorithm, one of the outstanding human abilities, includes a unimodal model, a multi-modal model, and a multi-modal method using music videos. In this study, the results were better by suggesting a system to analyze each song's spectrum into short samples and vote for the results. Among Deep Learning algorithms, CNN showed superior performance in the category of music genre compared to RNN, and improved performance when CNN and RNN were applied together. The system of voting for each CNN result by Deep Learning a short sample of music showed better results than the previous model and the model with Softmax layer added to the model performed best. The need for the explosive growth of digital media and the automatic classification of music genres in numerous streaming services is increasing. Future research will need to reduce the proportion of undifferentiated songs and develop algorithms for the last category classification of undivided songs.

• Journal of the Korea Society of Computer and Information
• /
• v.26 no.1
• /
• pp.57-67
• /
• 2021

Recently, multi-modal deep learning techniques that combine heterogeneous data for deep learning analysis have been utilized a lot. In particular, studies on the synthesis of Text to Image that automatically generate images from text are being actively conducted. Deep learning for image synthesis requires a vast amount of data consisting of pairs of images and text describing the image. Therefore, various data augmentation techniques have been devised to generate a large amount of data from small data. A number of text augmentation techniques based on synonym replacement have been proposed so far. However, these techniques have a common limitation in that there is a possibility of generating a incorrect text from the content of an image when replacing the synonym for a noun word. In this study, we propose a text augmentation method to replace words using word hierarchy information for noun words. Additionally, we performed experiments using MSCOCO data in order to evaluate the performance of the proposed methodology.