• Annual Conference on Human and Language Technology
• /
• 2015.10a
• /
• pp.115-118
• /
• 2015

Neural Machine Translation (NMT) 모델은 단일 신경망 구조만을 사용하는 End-to-end 방식의 기계번역 모델로, 기존의 Statistical Machine Translation (SMT) 모델에 비해서 높은 성능을 보이고, Feature Engineering이 필요 없으며, 번역 모델 및 언어 모델의 역할을 단일 신경망에서 수행하여 디코더의 구조가 간단하다는 장점이 있다. 그러나 NMT 모델은 출력 언어 사전(Target Vocabulary)의 크기에 비례해서 학습 및 디코딩의 속도가 느려지기 때문에 출력 언어 사전의 크기에 제한을 갖는다는 단점이 있다. 본 논문에서는 NMT 모델의 출력 언어 사전의 크기 제한 문제를 해결하기 위해서, 입력 언어는 단어 단위로 읽고(Encoding) 출력 언어를 문자(Character) 단위로 생성(Decoding)하는 방법을 제안한다. 출력 언어를 문자 단위로 생성하게 되면 NMT 모델의 출력 언어 사전에 모든 문자를 포함할 수 있게 되어 출력 언어의 Out-of-vocabulary(OOV) 문제가 사라지고 출력 언어의 사전 크기가 줄어들어 학습 및 디코딩 속도가 빨라지게 된다. 실험 결과, 본 논문에서 제안한 방법이 영어-일본어 및 한국어-일본어 기계번역에서 기존의 단어 단위의 NMT 모델보다 우수한 성능을 보였다.

• Proceedings of the KIEE Conference
• /
• 2001.07d
• /
• pp.2369-2373
• /
• 2001

Usually the static scheduling algorithms such as DMS (Deadline Monotonic Scheduling) or RMS(Rate Monotonic Scheduling) are used for CAN scheduling due to its ease with implementation. However, due to their inherently low utilization of network media, some dynamic scheduling approaches have been studied to enhance the utilization. In case of dynamic scheduling algorithms, two considerations are needed. The one is a priority inversion due to rough deadline encoding into stricted arbitration fields of CAN. The other is an arbitration delay due to the non-preemptive feature of CAN. In this paper, an extended algorithm is proposed from an existing EDS(Earliest Deadline Scheduling) approach of CAN scheduling algorithm haying a solution to the priority inversion. In the proposed algorithm, the available bandwidth of network media can be checked dynamically by all nodes. Through the algorithm, arbitration delay causing the miss of their deadline can be avoided in advance. Also non real-time messages can be processed with their bandwidth allocation. The proposed algorithm can achieve full network utilization and enhance aperiodic responsiveness, still guaranteeing the transmission of periodic messages.

• Journal of the Institute of Convergence Signal Processing
• /
• v.3 no.1
• /
• pp.41-47
• /
• 2002

Fractal image compression can reduce the size of image data by the contractive mapping that is affine transformation to find the block(called as range block) which is the most similar to the original image. Even though fractal image compression is regarded as an efficient way to reduce the data size, it has high distortion rate and requires long encoding time. In this paper, we presented a hybrid fractal image compression system with the modified SOFM Vector Quantizer which uses improved competitive learning method. The simulation results showed that the VQ hybrid fractal using improved competitive loaming SOFM has better distortion rate than the VQ hybrid fractal using normal SOFM.

• International Journal of Contents
• /
• v.17 no.4
• /
• pp.1-15
• /
• 2021

Named entity recognition (NER) is a crucial task for NLP, which aims to extract information from texts. To build NER systems, deep learning (DL) models are learned with dictionary features by mapping each word in the dataset to dictionary features and generating a unique index. However, this technique might generate noisy labels, which pose significant challenges for the NER task. In this paper, we proposed DL-dictionary features, and evaluated them on two datasets, including the OntoNotes 5.0 dataset and our new infectious disease outbreak dataset named GFID. We used (1) a Bidirectional Long Short-Term Memory (BiLSTM) character and (2) pre-trained embedding to concatenate with (3) our proposed features, named the Convolutional Neural Network (CNN), BiLSTM, and self-attention dictionaries, respectively. The combined features (1-3) were fed through BiLSTM - Conditional Random Field (CRF) to predict named entity classes as outputs. We compared these outputs with other predictions of the BiLSTM character, pre-trained embedding, and dictionary features from previous research, which used the exact matching and partial matching dictionary technique. The findings showed that the model employing our dictionary features outperformed other models that used existing dictionary features. We also computed the F1 score with the GFID dataset to apply this technique to extract medical or healthcare information.

• The Transactions of the Korea Information Processing Society
• /
• v.13 no.10
• /
• pp.513-519
• /
• 2024

Advancements in artificial intelligence(AI) have significantly improved deep learning-based audio deepfake technology, which has been exploited for criminal activities. To detect audio deepfake, we propose CoNSIST, an advanced audio deepfake detection model. CoNSIST builds on AASIST, which a graph-based end-to-end model, by integrating three key components: Squeeze and Excitation, Positional Encoding, and Reformulated HS-GAL. These additions aim to enhance feature extraction, eliminate unnecessary operations, and incorporate diverse information. Our experimental results demonstrate that CoNSIST significantly outperforms existing models in detecting audio deepfakes, offering a more robust solution to combat the misuse of this technology.

• Journal of the Institute of Electronics Engineers of Korea SP
• /
• v.44 no.2 s.314
• /
• pp.1-11
• /
• 2007

H.264/AVC uses variable block sizes to achieve significant coding gain. It has 7 different coding modes having different motion compensation block sizes in Inter slice, and 2 different intra prediction modes in Intra slice. This fine-tuned new coding feature has achieved far more significant coding gain compared with previous video coding standards. However, extremely high computational complexity is required when rate-distortion optimization (RDO) algorithm is used. This computational complexity is a major problem in implementing real-time H.264/AVC encoder on computationally constrained devices. Therefore, there is a clear need for complexity reduction algorithm of H.264/AVC such as fast mode decision. In this paper, we propose a fast mode decision with early $P8\times8$ mode rejection based on block size activity using large block history map (LBHM). Simulation results show that without any meaningful degradation, the proposed method reduces whole encoding time on average by 53%. Also the hybrid usage of the proposed method and the early SKIP mode decision in H.264/AVC reference model reduces whole encoding time by 63% on average.

• The KIPS Transactions:PartC
• /
• v.13C no.5 s.108
• /
• pp.541-548
• /
• 2006

Sensor network is a network for realizing the ubiquitous computing circumstances, which aggregates data by means of observation or detection deployed at the inaccessible places with the capacities of sensing and communication. To realize this circumstance, data which sensor nodes gathered from sensor networks are delivered to users, in which it is required to encrypt the data for the guarantee of secure communications. Therefore, it is needed to design key management scheme for encoding appropriate to the sensor nodes which feature continual data transfer, limited capacity of computation and storage and battery usage. We propose a key management scheme which is appropriate to sensor networks organizing hierarchical architecture. Because sensor nodes send data to their parent node, we can reduce routing energy. We assume that sensor nodes have different security levels by their levels in hierarchy. Our key management scheme provides different key establishment protocols according to the security levels of the sensor nodes. We reduce the number of sensor nodes which share the same key for encryption so that we reduce the damage by key exposure. Also, we propose key update protocols which take different terms for each level to update established keys efficiently for secure data encoding.

• Journal of KIISE:Computing Practices and Letters
• /
• v.8 no.6
• /
• pp.736-745
• /
• 2002

In network delivery of compressed video, packets may be lost if the channel is unreliable like Internet. Such losses tend to of cur in burst like continuous bit-stream error. In this paper, we propose an effective error-concealment approach to which an error resilient video encoding approach is applied against burst errors and which reduces a complexity of error concealment at the decoder using data hiding. To improve the performance of error concealment, a temporal and spatial error resilient video encoding approach at encoder is developed to be robust against burst errors. For spatial area of error concealment, block shuffling scheme is introduced to isolate erroneous blocks caused by packet losses. For temporal area of error concealment, we embed parity bits in content data for motion vectors between intra frames or continuous inter frames and recovery loss packet with it at decoder after transmission While error concealment is performed on error blocks of video data at decoder, it is computationally costly to interpolate error video block using neighboring information. So, in this paper, a set of feature are extracted at the encoder and embedded imperceptibly into the original media. If some part of the media data is damaged during transmission, the embedded features can be extracted and used for recovery of lost data with bi-direction interpolation. The use of data hiding leads to reduced complexity at the decoder. Experimental results suggest that our approach can achieve a reasonable quality for packet loss up to 30% over a wide range of video materials.

• Journal of Korea Multimedia Society
• /
• v.13 no.12
• /
• pp.1767-1777
• /
• 2010

In digital TV broadcasting, such as terrestrial, cable, satellite, and IPTV, the head-end of digital TV broadcasting has a more complicated transmission structure than that of analog TV broadcasting. Furthermore, digital TV broadcasting has a feature that supports multiplex models, such as Multiple Program Transport Stream (MPTS). Therefore, the purpose of our work was to design and examine a more efficient new system of emergency alert message transmission to support the digital TV broadcasting environments. Digital TV broadcasting is the IP generation or RF transmission of 8-VSB, QAM, and QPSK modulated through a multiplexer or re-multiplexer multiplexed stream as a MPEG-2 Transport Stream after content encoding. The new system proposed in this paper transmits an emergency alert message without scrambling after replacing the PID and payload of the -packet with the message prototype in the TS stream from the multiplexer. If we need to transmit an emergency alert message under digital TV broadcasting services, then the receiver first checks the PID of each packet in the TS stream for the emergency alert message. Next, if a packet is determined to be an emergency alert message, then the set-top box displays the message on the TV screen using its function of On Screen Display, or the PC based software displays the message on the monitor screen using its function of overlay with user interface if the packet is found to be an emergency alert message. We have designed an emergency alert message protocol and a system model. By experiments and analysis of the system, we concluded that the system achieved efficiency and the ability to send and receive emergency alert messages using the system under different digital TV broadcasting service environments.

• Science of Emotion and Sensibility
• /
• v.21 no.3
• /
• pp.29-48
• /
• 2018

Although the concept of "common sense" is often taken for granted, judging whether behavior or knowledge is common sense requires a complex series of mental processes. Additionally, different perceptions of common sense can lead to social conflicts. Thus, it is important to understand how we perceive common sense and make relevant judgments. The present study investigated the dynamics of neural representations underlying judgments of what common sense is. During functional magnetic resonance imaging, participants indicated the extent to which they thought that a given sentence corresponded to common sense under the given perspective. We incorporated two different decision contexts involving different cultural perspectives to account for social variability of the judgments, an important feature of common sense judgments apart from logical true/false judgments. Our findings demonstrated that common sense versus non-common sense perceptions involve the amygdala and a brain network for episodic memory recollection, including the hippocampus, angular gyrus, posterior cingulate cortex, and ventromedial prefrontal cortex, suggesting integrated affective, mnemonic, and social functioning in common sense processing. Furthermore, functional connectivity multivariate pattern analysis revealed that interactivity among the amygdala, angular gyrus, and parahippocampal cortex reflected representational features of common sense perception and not those of non-common sense perception. Our study demonstrated that the social memory network is exclusively involved in processing common sense and not non-common sense. These results suggest that intergroup exclusion and misunderstanding can be reduced by experiencing and encoding long-term social memories about behavioral norms and knowledge that act as common sense of the outgroup.