• Annual Conference on Human and Language Technology
• /
• 2021.10a
• /
• pp.410-413
• /
• 2021

본 논문에서는 분류 모델의 주류로 사용되고 있는 Encoder 기반 사전학습 모델(BERT, ALBERT, ELECTRA)의 내부 Encoder Layer가 하부 Layer에서는 Syntactic한 분석을 진행하고 상부 Layer로 갈수록 Semantic 한 분석을 진행하는 점, Layer가 구성됨에 따라 Semantic 정보가 Syntactic 정보를 개선해 나간다 점에 기반한 기존 연구 결과를 바탕으로 Encoder Layer를 구성함에 따라 어떻게 성능이 변화하는지 측정한다. 그리고 의도 분류를 위한 학습 데이터 셋도 분류하고자 하는 성격에 따라 Syntactic한 구성과 Semantic한 구성을 보인다는 점에 착안하여 ALBERT 및 ELECTRA를 이용한 의도 분류 모델을 구축하고 각 데이터 셋에 맞는 최적의 Encoder Layer 구성을 가지는 모델을 비교한 결과, 두 데이터 셋 간에 다른 Layer 구성을 보이는 점과 기존 모델보다 성능이 향상됨을 확인하였다.

• Proceedings of the IEEK Conference
• /
• 2002.07a
• /
• pp.50-53
• /
• 2002

There are increasing needs to deliver the multimedia streaming over heterogeneous networks. When considering network environments and equipment accessed by user, delivery of video streaming must be scalable. There are many kinds of scalable video coding: spatial, temporal, SNR, and hybrid. The SNR scalable and spatial resolution, but different SNR quality with respect to layers. The 1-layer SNR scalable encoder produces SNR scalable video streams with ease. But, there is drift problem. Modified 1-layer approach does not have this problem but coding inefficiency, and is not MPEG-compliant. The present MPEG-compliant 2-layer encoder comes out to reduce coding rate. But it still use only base layer to encode whole layer. In this paper, we propose adaptive MPEG-compliant 2-layer encoder. Using linear combination algorithm, encoder use 1 motion vector to encode the sequences efficiently. By dong this, we can achieve the coding efficiency of SNR scalable coding.

• Journal of Positioning, Navigation, and Timing
• /
• v.8 no.4
• /
• pp.193-200
• /
• 2019

The auto-encoder network which is a good candidate to handle the modeling of the signal strength attenuation is designed for denoising and compensating the distortion of the received data. It provides a non-linear mapping function by iteratively learning the encoder and the decoder. The encoder is the non-linear mapping function, and the decoder demands accurate data reconstruction from the representation generated by the encoder. In addition, the adaptive network width which supports the automatic generation of new hidden nodes and pruning of inconsequential nodes is also implemented in the proposed algorithm for increasing the efficiency of the algorithm. Simulation results show that the proposed method can improve the neural network training surface to achieve the highest possible accuracy of the signal modeling compared with the conventional modeling method.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• v.14 no.9
• /
• pp.3823-3840
• /
• 2020

Recently, auto-encoder has emerged as the most popular method in convolutional neural network (CNN) based image compression and has achieved impressive performance. In the traditional auto-encoder based image compression model, the encoder simply sends the features of last layer to the decoder, which cannot allocate bits over different spatial regions in an efficient way. Besides, these methods do not fully exploit the contextual information under different receptive fields for better reconstruction performance. In this paper, to solve these issues, a novel auto-encoder model is designed for image compression, which can effectively transmit the hierarchical features of the encoder to the decoder. Specifically, we first propose an adaptive bit-allocation strategy, which can adaptively select an importance channel. Then, we conduct the multiply operation on the generated importance mask and the features of the last layer in our proposed encoder to achieve efficient bit allocation. Moreover, we present an additional novel perceptual loss function for more accurate image details. Extensive experiments demonstrated that the proposed model can achieve significant superiority compared with JPEG and JPEG2000 both in both subjective and objective quality. Besides, our model shows better performance than the state-of-the-art convolutional neural network (CNN)-based image compression methods in terms of PSNR.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• v.18 no.7
• /
• pp.1706-1725
• /
• 2024

The unique U-shaped structure of U-Net network makes it achieve good performance in image segmentation. This network is a lightweight network with a small number of parameters for small image segmentation datasets. However, when the medical image to be segmented contains a lot of detailed information, the segmentation results cannot fully meet the actual requirements. In order to achieve higher accuracy of medical image segmentation, a novel improved U-Net network architecture called multi-scale encoder-decoder U-Net+ (MEDU-Net+) is proposed in this paper. We design the GoogLeNet for achieving more information at the encoder of the proposed MEDU-Net+, and present the multi-scale feature extraction for fusing semantic information of different scales in the encoder and decoder. Meanwhile, we also introduce the layer-by-layer skip connection to connect the information of each layer, so that there is no need to encode the last layer and return the information. The proposed MEDU-Net+ divides the unknown depth network into each part of deconvolution layer to replace the direct connection of the encoder and decoder in U-Net. In addition, a new combined loss function is proposed to extract more edge information by combining the advantages of the generalized dice and the focal loss functions. Finally, we validate our proposed MEDU-Net+ MEDU-Net+ and other classic medical image segmentation networks on three medical image datasets. The experimental results show that our proposed MEDU-Net+ has prominent superior performance compared with other medical image segmentation networks.

• Journal of Broadcast Engineering
• /
• v.2 no.1
• /
• pp.8-15
• /
• 1997

In this paper, a real-time operating Motion Picture Experts Group-1 (MPEG-1) audio encoder system is implemented using a TMS320C31 Digital Signal Processor (DSP) chip. The basic operation of the MPEG-1 audio encoder algorithm based on audio layer-2 and psychoacoustic model-1 is first verified by C-language. It is then realized using the Texas Instruments (Tl) assembly in order to reduce the overall execution time. Finally, the actual BSP circuit board for the encoder system is designed and implemented. In the system, the side-modules such as the analog-to-digital converter (ADC) control, the input/output (I/O) control, the bit-stream transmission from the DSP board to the PC and so on, are utilized with a field programmable gate array (FPGA) using very high speed hardware description language (VHDL) codes. The complete encoder system is able to process the stereo audio signal in real-time at the sampling frequency 48 kHz, and produces the encoded bit-stream with the bit-rate 192 kbps. The real-time operation capability of the encoder system and the good quality of the decoded sound are also confirmed using various types of actual stereo audio signals.

• Journal of Korea Multimedia Society
• /
• v.22 no.11
• /
• pp.1251-1258
• /
• 2019

Intelligent tutoring system enables users to effectively learn by utilizing various artificial intelligence techniques. For instance, it can recommend a proper curriculum or learning method to individual users based on their learning history. To do this effectively, user's characteristics need to be analyzed and classified based on various aspects such as interest, learning ability, and personality. Even though data labeled by the characteristics are required for more accurate classification, it is not easy to acquire enough amount of labeled data due to the labeling cost. On the other hand, unlabeled data should not need labeling process to make a large number of unlabeled data be collected and utilized. In this paper, we propose a semi-supervised learning method based on feedback variational auto-encoder(FVAE), which uses both labeled data and unlabeled data. FVAE is a variation of variational auto-encoder(VAE), where a multi-layer perceptron is added for giving feedback. Using unlabeled data, we train FVAE and fetch the encoder of FVAE. And then, we extract features from labeled data by using the encoder and train classifiers with the extracted features. In the experiments, we proved that FVAE-based semi-supervised learning was superior to VAE-based method in terms with accuracy and F1 score.

• Proceedings of the Korea Information Processing Society Conference
• /
• 2022.11a
• /
• pp.905-907
• /
• 2022

악성코드를 분석하는 기존 기법인 정적분석은 빠르고 효율적으로 악성코드를 탐지할 수 있지만 난독화된 파일에 취약한 반면,, 동적분석은 난독화된 파일에 적합하지만 느리고 비용이 많이 든다는 단점을 가진다. 본 연구에서는 두 분석 기법의 단점을 해결하기 위해 딥러닝 모델을 활용한 난독화에 강한 정적분석 모델을 제안하였다. 본 연구에서 제안한 방법은 원본 코드 및 난독화된 파일을 grayscale 이미지로 변환하여 데이터셋을 구축하고 AutoEncoder 를 사전학습시켜 encoder 가 원본 파일과 난독화된 파일로부터 원본 파일의 특징을 추출할 수 있도록 한 이후, encoder 의 output 을 fully connected layer 의 입력으로 넣고 전이학습시켜 악성코드를 탐지하도록 하였다. 본 연구에서는 제안한 방법론은 난독화된 파일에서 악성코드를 탐지하는 성능을 F1 score 기준 14.17% 포인트 향상시켰고, 난독화된 파일과 원본 파일을 전체를 합친 데이터셋에서도 악성코드 탐지 성능을 F1 score 기준 7.22% 포인트 향상시켰다.

• Journal of the Institute of Electronics and Information Engineers
• /
• v.50 no.6
• /
• pp.260-264
• /
• 2013

This paper proposes a fast enhancement layer coding method to reduce computational complexity for Scalable HEVC (SHVC) which is based on High Efficiency Video Coding (HEVC). The proposed method decreases encoding time by simplifying Rate Distortion Optimization (RDO)for enhancement layers (EL). The simplification is achieved by restricting CU depths based on the correlation of coding unit (CU) depths between adjacent layers and scalability (spatial or quality) of EL. Comparing with the performance of SHM 1.0 software encoder, the proposed method reduces the encoding time by up to 31.5%.

• Proceedings of the Korean Society of Computer Information Conference
• /
• 2022.07a
• /
• pp.23-26
• /
• 2022

딥 러닝 모델은 블랙 박스 (Black Box) 모델로 예측에 대한 근거를 제시하지 못해 신뢰성이 떨어지는 단점이 존재한다. 이를 해결하기 위해 딥 러닝 모델에 설명력을 부여하는 설명 가능한 인공지능 (XAI) 분야 연구가 활발하게 이루어지고 있다. 본 논문에서는 모델 예측을 프로토타입을 통해 설명하는 딥 러닝 모델을 제시한다. 즉, "주어진 이미지는 티셔츠인데, 그 이유는 티셔츠를 대표하는 모양의 프로토타입과 닮았기 때문이다."의 형태로 딥 러닝 모델을 설명한다. 해당 모델은 Encoder, Prototype Layer, Classifier로 구성되어 있다. Encoder는 Feature를 추출하는 데 활용하고 Classifier를 통해 분류 작업을 수행한다. 모델이 제시하는 분류 결과를 설명하기 위해 Prototype Layer에서 가장 유사한 프로토타입을 찾아 설명을 제시한다. 실험 결과 프로토타입 생성 기반 설명 모델은 기존 이미지 분류 모델과 유사한 예측 정확도를 보였고, 예측에 대한 설명력까지 확보하였다.