This paper presents both 1×1 and 3×3 convolution accelerators for convolutional neural networks. Both the 1×1 convolution accelerator and the 3×3 convolution accelerator process 16 input channels and 1 output channel per unit time. The 1×1 convol...
http://chineseinput.net/에서 pinyin(병음)방식으로 중국어를 변환할 수 있습니다.
변환된 중국어를 복사하여 사용하시면 됩니다.
https://www.riss.kr/link?id=A109202852
2024
Korean
569
학술저널
1768-1771(4쪽)
0
상세조회0
다운로드다국어 초록 (Multilingual Abstract)
This paper presents both 1×1 and 3×3 convolution accelerators for convolutional neural networks. Both the 1×1 convolution accelerator and the 3×3 convolution accelerator process 16 input channels and 1 output channel per unit time. The 1×1 convol...
This paper presents both 1×1 and 3×3 convolution accelerators for convolutional neural networks. Both the 1×1 convolution accelerator and the 3×3 convolution accelerator process 16 input channels and 1 output channel per unit time. The 1×1 convolution accelerator processes 18 pixels, while the 3×3 convolution accelerator processes 2 pixels. These two accelerators are designed with an identical structure, allowing the simultaneous utilization of multipliers and adders. The designed convolution accelerator utilizes a total of 288 multipliers and 286 adders, along with 2W+9 registers for the width W of the feature map.
디지털 트윈 내 복합추론을 위한 협력적 모델 추론 방식
디지털 트윈 연합을 위한 트윈 데이터 정제 플랫폼 프로토타입 구현
한의건강검진 자료 기반 대사증후군 위험집단 비침습적 예측모형 개발