• 제목/요약/키워드: SoftMax

검색결과 102건 처리시간 0.02초

CNN의 SoftMax 연산을 위한 연속 근사 방식의 로그 연산 회로 (Successive Approximated Log Operation Circuit for SoftMax in CNN)

  • Kang, Hyeong-Ju
    • 한국정보통신학회논문지
    • /
    • 제25권2호
    • /
    • pp.330-333
    • /
    • 2021
  • In a CNN for image classification, a SoftMax layer is usually placed at the end. The exponentinal and logarithmic operations in the SoftMax layer are not adequate to be implemented in an accelerator circuit. The operations are usually implemented with look-up tables, and the exponential operation can be implemented in an iterative method. This paper proposes a successive approximation method to calculate a logarithm to remove a very large look-up table. By substituing the large table with two very small tables, the circuit can be reduced much. The experimental results show that the 85% area reduction can be reached with a small error degradation.

CNN에서 입력 최댓값을 이용한 SoftMax 연산 기법 (SoftMax Computation in CNN Using Input Maximum Value)

  • Kang, Hyeong-Ju
    • 한국정보통신학회논문지
    • /
    • 제26권2호
    • /
    • pp.325-328
    • /
    • 2022
  • A convolutional neural network(CNN) is widely used in the computer vision tasks, but its computing power requirement needs a design of a special circuit. Most of the computations in a CNN can be implemented efficiently in a digital circuit, but the SoftMax layer has operations unsuitable for circuit implementation, which are exponential and logarithmic functions. This paper proposes a new method to integrate the exponential and logarithmic tables of the conventional circuits into a single table. The proposed structure accesses a look-up table (LUT) only with a few maximum values, and the LUT has the result value directly. Our proposed method significantly reduces the space complexity of the SoftMax layer circuit implementation. But our resulting circuit is comparable to the original baseline with small degradation in precision.

Maximum shear modulus of rigid-soft mixtures subjected to overconsolidation stress history

  • Boyoung Yoon;Hyunwook Choo
    • Geomechanics and Engineering
    • /
    • 제37권5호
    • /
    • pp.443-452
    • /
    • 2024
  • The use of sand-tire chip mixtures in construction industry is a sustainable and environmentally friendly approach that addresses both waste tire disposal and soil improvement needs. However, the addition of tire chip particles to natural soils decreases maximum shear modulus (Gmax), but increases compressibility, which can be potential drawbacks. This study examines the effect of overconsolidation stress history on the maximum shear modulus (Gmax) of rigid-soft mixtures with varying size ratios (SR) and tire chip contents (TC) by measuring the wave velocity through a 1-D compression test during loading and unloading. The results demonstrate that the Gmax of tested mixtures in the normally consolidated state increased with increasing SR and decreasing TC. However, the tested mixtures with a smaller SR exhibited a greater increase in Gmax during unloading because of the active pore-filling behavior of the smaller rubber particles and the consequent increased connectivity between sand particles. The SR-dependent impact of the overconsolidation stress history on Gmax was verified using the ratio between the swelling and compression indices. Most importantly, this study reveals that the excessive settlement and lower Gmax of rigid-soft mixtures can be overcome by introducing an overconsolidated state in sand-tire chip mixtures with low TC.

Max-Log-MAP을 이용한 Gray 부호화된 PAM 신호의 연판정 계산식 (Soft decision for Gray Coded PAM Signals Using Max-Log-MAP)

  • 현광민;윤동원
    • 한국통신학회논문지
    • /
    • 제31권2C호
    • /
    • pp.117-122
    • /
    • 2006
  • 본 논문에서는 로그 최우비(log likelihood ratio, LLR)를 이용하여 Gray 부호화된 PAM신호를 위한 비트 연판정 계산식을 제안한다. 이 계산식은 Gray 매핑 특성을 이용하여 Max-Log-MAP 알고리듬에서 필요한 max0/min0 함수를 사용하지 않고 산술 연산만을 사용하기 때문에 구현이 간단하다. 제안된 식의 결과는 기존 Max-Log-MAP 알고리듬의 결과와 일치한다. 또한, 식에 사용되는 인자들은 송수신 시스템이 서로 공유하는 정보와 수신된 심벌 값만을 이용하여 계산한다. 따라서 본 논문에서 제안된 알고리듬은 일반적으로 많이 사용하는 이진 반복 복호기 등과 함께 실제 응용 설계에 적용이 가능하며, 특히 실제 설계에 적용되는 기존의 여러 가지 알고리듬에 비하여 구조가 유연하고 효율적이며 정확한 비트별 LLR을 제공하는 효율적인 방법 중의 하나이다.

A New Generation of Soft X-Ray Lasers

  • Janulewicz, K.A.;Tuemmler, J.;Stiel, H.;Nickles, P.V.
    • 한국광학회:학술대회논문집
    • /
    • 한국광학회 2007년도 동계학술발표회 논문집
    • /
    • pp.319-320
    • /
    • 2007
  • PDF

F-18 FDG PET/CT에서 양성과 악성 근골격 종양의 감별진단 - 수신자 판단특성곡선을 이용한 maxSUV의 절단값 결정 (The Differentiation of Malignant and Benign Musculoskeletal Tumors by F-18 FDG PET/CT Studies - Determination of maxSUV by Analysis of ROC Curve)

  • 공은정;조인호;천경아;원규장;이형우;최준혁;신덕섭
    • Nuclear Medicine and Molecular Imaging
    • /
    • 제41권6호
    • /
    • pp.553-560
    • /
    • 2007
  • 목적: F-18 FDG PET은 근골격계 종양에서 양성과 악성병변을 감별하는 유용성에 대하여 다양한 결과가 보고되고 있다. 저자들은 F-18 FDG를 이용한 PET/CT로 근골격계 종양의 maxSUV를 분석하고 비교하여 유용성을 알아보았다. 대상 및 방법: 치료 전 46개 병소(연부 조직 종양 양성/악성 : 11/12, 골종양 양성/악성 9/14)에 대하여 F-18 FDG PET/CT를 시행하였으며, 조직학적 검사로 확진하였다. 악성과 양성을 구분하는 maxSUV 절단값은 연부 조직 종양에서는 4.1, 골종양에서는 3.05로 하였다. 결과: 연부 조직 종양에서 양성(R=11; maxSUV $3.4{\pm}3.2$)과 악성(n=12; maxSUV $14.8{\pm}12.2$) 간에 maxSUV는 통계학적으로 유의하게 (p<0.001) 차이가 있었다. 민감도와 특이도는 각각 83%, 91%였다. 그러나 골종양에서는 양성 종양(n=9; maxSUV $5.4{\pm}4.0$)과 악성 골종양(n=14; maxSUV $7.3{\pm}3.2$) 간에 통계학적으로 유의한 차이를 보이지 않았다. 연부 조직 종양에서는 결절성 근막염이 위양성으로 나타났고(maxSUV=12.4) 골종양에서는 섬유성 골이형성증과 랑게르한스세포 조직구증식증 2예 및 골모세포종이 있었다. 결론: 연부 조직 종양에서 maxSUV는 양성과 악성을 감별하는데 유용하였다. 그러나 골종양의 경우에는 maxSUV가 낮은 경우에는 악성을 배제할 수 있었으나, maxSUV가 높은 경우에는 조직학적으로 조직구나 섬유모세포 등이 포함된 종양의 감별진단을 고려하여야한다.

Bioequivalence of Cholicerin Soft Capsule to Gliatilin Soft Capsule (Choline Alphoscerate 400 mg)

  • Kang, Hyun-Ah;Kim, Se-Mi;Kang, Seung-Rae;Kang, Min-Sun;Lee, Sang-No;Kwon, In-Ho;Yoo, Hee-Doo;Kim, Yoon-Gyoon;Lee, Yong-Bok
    • Journal of Pharmaceutical Investigation
    • /
    • 제40권2호
    • /
    • pp.109-115
    • /
    • 2010
  • The purpose of the present study was to evaluate the bioequivalence of two choline alphoscerate soft capsules, Gliatilin soft capsule (Daewoong Pharmaceuticals Co., Ltd.) and Cholicerin soft capsule (Sam Chun Dang Pharm. Co., Ltd.), according to the guidelines of Korea Food and Drug Administration (KFDA). Serum concentrations of choline after oral administration of choline alphoscerate were determined using a validated LC/MS/MS method. This method showed linear response over the concentration range of 0.5-20 ${\mu}g$/mL with correlation coefficient of 0.9999. The lower limit of quantitation using 100 ${\mu}L$ of serum was 0.5 ${\mu}g$/mL which was sensitive enough for pharmacokinetic studies. Thirty six healthy male Korean volunteers received each medicine at the choline alphoscerate dose of 1200 mg in a $2{\times}2$ crossover study. There was a one-week washout period between the doses. Blood samples were taken at predetermined time intervals up to 8 hr. $AUC_t$ (the area under the serum concentration-time curve from time 0 to 8 hr) was calculated by the linear trapezoidal rule method. $C_{max}$ (the maximum serum drug concentration) and $T_{max}$ (the time to reach $C_{max}$) were compiled from the serum concentration-time data. Analysis of variance was carried out using logarithmically transformed $AUC_t$ and $C_{max}$. No significant sequence effect was found for all of the bioavailability parameters, indicating that the crossover design was properly performed. The 90% confidence intervals of the $AUC_t$ ratio and the $C_{max}$ ratio for Cholicerin/Gliatilin were log0.9998-log1.1172 and log0.9938-1.0944, respectively. These values were within the acceptable bioequivalence intervals of log0.80-log1.25. Thus, the criteria of the KFDA guidelines for the bioequivalence was satisfied, indicating Cholicerin soft capsule and Gliatilin soft capsule are bioequivalent.

Gray 부호화된 QAM 신호를 위한 근사화된 MAP 알고리듬 (Approximated MAP Algorithm for Gray Coded QAM Signals)

  • 현광민
    • 한국산학기술학회논문지
    • /
    • 제10권12호
    • /
    • pp.3702-3707
    • /
    • 2009
  • 본 논문은 Gray 부호화된 QAM (Quadrature Amplitude Modulation) 신호를 I 축 상에서 M개의 심벌을 갖는 M-PAM (Pulse Amplitude Modulation)과 Q 축 상에서 N개의 심벌을 갖는 N-PAM으로 분리한다. 수신된 심벌 신호를 비트 연판정 값으로 변환하기 위하여 Euclidean 거리를 이용한 근사화된 MAP (Maximum a Posteriori) 알고리듬을 제시한다. 기존의 Max-Log-MAP 방식은 일반 MAP 방식에서 사용하는 지수함수 혹은 로그함수 대신 심벌간 거리 비교를 통하여 구현 복잡도를 낮추었다. 그러나 심벌의 수가 증가 할수록 비교대상이 많아지므로 구현 복잡도가 증가하게 된다. 제안된 알고리듬은 사칙 연산에 의해 계산이 되기 때문에 직관적으로 구현복잡도가 낮아짐을 알 수 있다.

A Receiver-Aided Seamless And Smooth Inter-RAT Handover At Layer-2

  • Liu, Bin;Song, Rongfang;Hu, Haifeng
    • KSII Transactions on Internet and Information Systems (TIIS)
    • /
    • 제9권10호
    • /
    • pp.4015-4033
    • /
    • 2015
  • The future mobile networks consist of hyper-dense heterogeneous and small cell networks of same or different radio access technologies (RAT). Integrating mobile networks of different RATs to provide seamless and smooth mobility service will be the target of future mobile converged network. Generally, handover from high-speed networks to low-speed networks faces many challenges from application perspective, such as abrupt bandwidth variation, packet loss, round trip time variation, connection disruption, and transmission blackout. Existing inter-RAT handover solutions cannot solve all the problems at the same time. Based on the high-layer convergence sublayer design, a new receiver-aided soft inter-RAT handover is proposed. This soft handover scheme takes advantage of multihoming ability of multi-mode mobile station (MS) to smooth handover procedure. In addition, handover procedure is seamless and applicable to frequent handover scenarios. The simulation results conducted in UMTS-WiMAX converged network scenario show that: in case of TCP traffics for handover from WiMAX to UMTS, not only handover latency and packet loss are eliminated completely, but also abrupt bandwidth/wireless RTT variation is smoothed. These delightful features make this soft handover scheme be a reasonable candidate of mobility management for future mobile converged networks.

Efficient Method to Implement Max-Log-MAP Algorithm: Parallel SOVA

  • 이창우
    • 한국통신학회논문지
    • /
    • 제33권6C호
    • /
    • pp.438-443
    • /
    • 2008
  • The efficient method to implement the Max-Log-MAP algorithm is proposed by modifying the conventional algorithm. It is called a parallel soft output Viterbi algorithm (SOVA) and the rigorous proof is given for the equivalence between the Max-Log-MAP algorithm and the parallel SOVA. The parallel SOVA is compared with the conventional algorithms and we show that it is an efficient algorithm implementing the modified SOVA in parallel.