• Annual Conference on Human and Language Technology
• /
• 2022.10a
• /
• pp.473-476
• /
• 2022

문장 임베딩은 문장의 의미를 고려하여 모델이 적절하게 의미적인 벡터 공간에 표상하는 것이다. 문장 임베딩을 위해 다양한 방법들이 제안되었지만, 최근 가장 높은 성능을 보이는 방법은 대조적 학습 방법이다. 대조적 학습을 이용한 문장 임베딩은 문장의 의미가 의미적으로 유사하면 가까운 공간에 배치하고, 그렇지 않으면 멀게 배치하도록 학습하는 방법이다. 이러한 대조적 학습은 비지도와 지도 학습 방법이 존재하는데, 본 논문에서는 효과적인 비지도 학습방법을 제안한다. 기존의 비지도 학습 방법은 문장 표현을 학습하는 언어모델이 자체적인 정보를 활용하여 문장의 의미를 구별한다. 그러나, 하나의 모델이 판단하는 정보로만 문장 표현을 학습하는 것은 편향적으로 학습될 수 있기 때문에 한계가 존재한다. 따라서 본 논문에서는 Cross-Encoder의 Re-Ranker를 통한 의미 검색으로부터 추천된 문장 쌍을 학습하여 기존 모델의 성능을 개선한다. 결과적으로, STS 테스크에서 베이스라인보다 2% 정도 더 높은 성능을 보여준다.

• Annual Conference on Human and Language Technology
• /
• 2020.10a
• /
• pp.188-191
• /
• 2020

본 연구는 COVID-19 질의 응답 태스크를 위한 Poly-encoder 기반의 태스크를 제안하였다. COVID-19 질의 응답 시스템은 사람들에게 최신 정보에 대해 빠르고 신뢰성이 높은 정보를 전달하는 특성을 가져야한다. 검색 기반 질의 응답 시스템은 pairwise 연산을 기반으로 수행되는데, Poly-encoder는 사전 학습된 트랜스포머(transformer)기반의 pairwise 연산 방법론 중 기존 Cross-encoder와 Bi-encoder보다 실사용 및 성능이 뛰어남을 보였다 [1]. 특히, Poly-encoder는 정확도가 높으면서도 빠른 응답속도를 가지며 검색기반의 각종 태스크에서 좋은 성능을 보였다. 따라서 본 연구는 COVID-19를 위한 Poly-encoder기반의 질의 응답 태스크를 위하여 기존 질의 응답 태스크와 페르소나 기반의 질의 응답 태스크로 두 가지 유형의 태스크를 생성하여 모델을 학습하였다. 또한 신뢰성 있는 리소스정보로부터 모델에 최신 정보 반영을 위하여 자동 크롤러를 구축하여 데이터를 수집하였다. 마지막으로 전문가를 통한 데이터셋을 구축하여 질문-응답과 질의어-질문에 대한 모델 검증을 수행하였다.

• Journal of the Institute of Electronics Engineers of Korea SP
• /
• v.47 no.3
• /
• pp.87-95
• /
• 2010

Portable multi-media products which can service the highest audio-quality by using lossless audio codec has been released and the international lossless codecs, MPEG-4 audio lossless coding(ALS) and MPEG-4 scalable lossless coding(SLS), were standardized by MPEG in 2006. The simple profile of MPEG-4 ALS, it supports up to stereo, was defined by MPEG in 2009. The lossless audio codec should have low-complexity in stereo to be widely used in portable multi-media products. But the previous researches of MPEG-4 ALS have focused on an improvement of compression ratio, a complexity reduction in multi-channels coding, and a selection of linear prediction coefficients(LPCs) order. In this paper, the complexity and compression ratio of MPEG-4 ALS encoder is analyzed in simple profile of MPEG-4 ALS, the method to reduce a complexity of MPEG-4 ALS encoder is proposed. Based on an analysis of complexity of MPEG-4 ALS encoder, the complexity of short-term prediction filter of MPEG-4 ALS encoder is reduced by using the low-complexity filter that is proposed in previous research to reduce the complexity of MPEG-4 ALS decoder. Also, we propose a joint coding decision method, it reduces the complexity and keeps the compression ratio of MPEG-4 ALS encoder. In proposed method, the operation of joint coding is decided based on the relation between cross-correlation of residual and compression ratio of joint coding. The performance of MPEG-4 ALS encoder that has the method and low-complexity filter is evaluated by using the MPEG-4 ALS conformance test file and normal music files. The complexity of MPEG-4 ALS encoder is reduced by about 24% by comparing with MPEG-4 ALS reference encoder, while the compression ratio by the proposed method is comparable to MPEG-4 ALS reference encoder.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• v.18 no.1
• /
• pp.1-14
• /
• 2024

Time-series forecasting is extensively used in the actual world. Recent research has shown that Transformers with a self-attention mechanism at their core exhibit better performance when dealing with such problems. However, most of the existing Transformer models used for time series prediction use the traditional encoder-decoder architecture, which is complex and leads to low model processing efficiency, thus limiting the ability to mine deep time dependencies by increasing model depth. Secondly, the secondary computational complexity of the self-attention mechanism also increases computational overhead and reduces processing efficiency. To address these issues, the paper designs an efficient multi-layer attention-based time-series forecasting model. This model has the following characteristics: (i) It abandons the traditional encoder-decoder based Transformer architecture and constructs a time series prediction model based on multi-layer attention mechanism, improving the model's ability to mine deep time dependencies. (ii) A cross attention module based on cross attention mechanism was designed to enhance information exchange between historical and predictive sequences. (iii) Applying a recently proposed sparse attention mechanism to our model reduces computational overhead and improves processing efficiency. Experiments on multiple datasets have shown that our model can significantly increase the performance of current advanced Transformer methods in time series forecasting, including LogTrans, Reformer, and Informer.

• The Journal of Korean Institute of Communications and Information Sciences
• /
• v.24 no.6B
• /
• pp.1134-1141
• /
• 1999

In this paper, we implement an optical code division multiple access(OCDMA) system and analyze the performance of the implemented system. In the implemented system, a transmitter encodes input data into optical pulses by using laser diode, and spreads the encoded pulses in an encoder which consists of 4 stage delay lines. The decoder which is the same structure as that of encoder delays and combines the spreaded pulses, and thus recovers the original data. At first, we discuss the auto- and cross-correlations of OCDMA signals under both environments of single user and multi-users, and then verify the simulation results with experimental results. We also evaluate the effect of a number of stages of delay line and code length on the system performance through computer simulations. As experimental results we can see that if the decoder have the same configuration as that of encoder, the peak auto-correlation characteristics can be achieved, and thus we can recover the original data from received data. As simulation results we can see that although bit error rate decreases as code length decreases or the number of stage of delay line increases, it is difficult to implement the system because the pulse width becomes narrow. From the results, we can apply CDMA technologies to optical communication networks.

• Journal of the Institute of Electronics Engineers of Korea SP
• /
• v.46 no.3
• /
• pp.119-124
• /
• 2009

In this paper, AMR-WB+ audio coder is implemented, in real-time, using Intel 400MHz Xscale PXA250 with 32bit RISC processor ARM9E-J(R)core. The assembly code for ARM9E-J(R)core is developed through the serial process of C code optimization, cross compile, assembly code manual optimization and adjusting the optimized code to Embedded Visual C++ platform. C code is trimmed on Visual C++ platform. Cross compile and assembly code manual optimization are performed on CodeWarrior with ARM compiler. Through these stages the code for both ARM EVM board and PDA is implemented. The average complexities of the code are 160.75MHz on encoder and 33.05MHz on decoder. In case of static link library(SLL), the required memories are 65.21Kbyte, 32.01Kbyte and 279.81Kbyte on encoder, decoder and common sources, respectively. The implemented coder is evaluated using 16 test vectors given by 3GPP to verify the bit-exactness of the coder.

• Journal of The Institute of Information and Telecommunication Facilities Engineering
• /
• v.2 no.4
• /
• pp.5-16
• /
• 2003

We have analyzed the performance of optical matched filters in the fiber-optic code division multiple access (FO-CDMA) system based on optical parallel coupler access encoder (PCAE) and parallel coupler access decoder (PCAD) by experiment. In previous studies, the performance evaluation of the FO-CDMA system using SCAE and SCAD was not sufficiently accurate because they analyzed system performance only considering the first order signals. Since optical SCAE and SCAD intrinsically have high order signals of various patterns as the number of coupler increases, they change auto- and cross-correlation intensities. Thus, it is necessary to investigate properties of the PCAE and PCAD so that we may analyze the exact performance of system. In this paper, it is found that the peak to sidelobe ratio using the PCAE and PCAD increases as $\alpha$ (coupling coefficient) value increases. Also, we found that the proposed PCAE and PCAD are superior to SCAE and SCAD in performance improvement.

• Journal of Internet Computing and Services
• /
• v.11 no.2
• /
• pp.49-60
• /
• 2010

H.264/AVC offers a better encoding efficiency than conventional video standards by adopting many new encoding techniques. However, the advanced coding techniques also add to the overall complexity for H.264/AVC encoder. Accordingly, it is necessary to perform optimization to alleviate the level of complexity for the video encoder. The amount of computation for motion estimation is of particular importance. In this paper, we propose a diamond web-grid search algorithm combined with efficient stationary block skip method which employs full diamond and dodecagon search patterns, and the variable thresholds are used for performing an effective skip of stationary blocks. The experimental results indicate that the proposed technique reduces the computations of the unsymmetrical-cross multi-hexagon-grid search algorithm by up to 12% while maintaining a similar PSNR performance.

• Korean Journal of Optics and Photonics
• /
• v.33 no.3
• /
• pp.106-112
• /
• 2022

A two-dimensional optical code division multiple access (OCDMA) encoder/decoder, which is composed of add/drop filters and all-pass filters for delay operation, is proposed. An example design is presented, and its feasibility is illustrated through numerical simulations. The chip area of the proposed OCDMA encoder/decoder could be about one-third that of a previous OCDMA device employing delay waveguides. Its performance is numerically investigated using the transfer-matrix method combined with the fast Fourier transform. The autocorrelation peak level over the maximum cross-correlation level for incorrect wavelength hopping and spectral phase code combinations is greater than 3 at the center of the correctly decoded pulse, which assures a bit error rate lower than 10^-3, corresponding to the forward error-correction limit.

• Journal of the Korea Society for Simulation
• /
• v.24 no.3
• /
• pp.61-68
• /
• 2015

Mobile ad hoc networks (MANETs) are self-organized dynamic networks populated by mobile nodes. This paper presents the improved cross-layer approach to complement the recent works for video transmission services on MANET. We use a statistical design of experiment and analysis in order to investigate interactions between major factors of each layer effectively with minimizing ns-3 simulation run time. The proposed cross-layer approach considers MANET protocol layers (i.e., physical, network and transmission layers) and an application layer (i.e., a video encoder) as factors simultaneously. In addition, the approach defines an objective video quality metric as a response variable. The result of this paper can be applicable as a preliminary research to design an optimized video transmission application which has ability to adjust controllable factors to dynamic uncontrollable factors.