没有合适的资源？快使用搜索试试~ 我知道了~

文库首页人工智能深度学习An Energy-Efficient Reconfigurable Accelerator for Deep CNN

An Energy-Efficient Reconfigurable Accelerator for Deep CNN

Eyeriss

4星 · 超过85%的资源需积分: 50 29 下载量 112 浏览量 2017-11-11 16:43:06 上传评论收藏 3.8MB PDF 举报

温馨提示

试读

43页

关于Eyeriss项目的介绍性文档，内容包括介绍卷积神经网络，介绍Eyeriss项目的基本架构、算法单元结构、数据流压缩方式、计算流程方式等。Eysriss属于MIT的一个项目，目前该项目已经流片。它的运行效率是普通手机GPU的10倍，因此可以直接在移动设备上执行强大的人工智能算法，而不需要将数据上传到互联网进行处理。

资源详情

资源评论

14.5: Eyeriss: An Energy-Efficient Reconfigurable Accelerator for Deep Convolutional Neural Networks

International Solid-State Circuits Conference

1of 43

Eyeriss: An Energy-Efficient

Reconfigurable Accelerator

for Deep Convolutional

Neural Networks

Yu-Hsin Chen

, Tushar Krishna

Joel Emer

1, 2

, Vivienne Sze

MIT

NVIDIA

14.5: Eyeriss: An Energy-Efficient Reconfigurable Accelerator for Deep Convolutional Neural Networks

International Solid-State Circuits Conference

2of 43

Future of Deep Learning

Recognit ion

Sel f -Driving Cars

DCNN Accelerator is Crucial

• High Throughput for Real-time

Processing

• Sub-watt Power/Energy

Consumption

DPM

Accuracy

0.1k – 0.5k OP/Px

DCNN

15k – 300k OP/Px

Computation

• DPM: Deformable Part Model

14.5: Eyeriss: An Energy-Efficient Reconfigurable Accelerator for Deep Convolutional Neural Networks

International Solid-State Circuits Conference

3of 43

DCNN Explained

Convolution

Activation

Norm.

Pooling

Modern Deep CNN: 5 – 152 Layers

Classes

Conv

Layer

Conv

Layer

Low-Level

Features

High-Level

Features

14.5: Eyeriss: An Energy-Efficient Reconfigurable Accelerator for Deep Convolutional Neural Networks

International Solid-State Circuits Conference

4of 43

Convolution is the Most Important

Activation

Norm.

Pooling

Modern Deep CNN: 5 – 152 Layers

Classes

Conv

Layer

Conv

Layer

Low-Level

Features

High-Level

Features

Convolution

Takes 90% – 99% of

Computation and Runtime

14.5: Eyeriss: An Energy-Efficient Reconfigurable Accelerator for Deep Convolutional Neural Networks

International Solid-State Circuits Conference

5of 43

Convolution in CNN

Input Image

Output Image

Filter

剩余42页未读，继续阅读

评论收藏

内容反馈

yanxiaopan

2020-02-10

不是论文，只是简单的PPt介绍，不值14分

An Energy-Efficient Reconfigurable Accelerator for Deep CNN

评论2

最新资源

An Energy-Efficient Reconfigurable Accelerator for Deep CNN

评论2

最新资源

相关推荐

Accelerator

An FPGA-Based Resource-Saving Hardware Accelerator for Deep Neural Network.pdf

Eyeriss v1 + v2 论文

eyeriss项目组的深度学习加速器的总结

基于FPGA的深度学习加速器设计与实现

MIT深度神经网络硬件架构设计教程

PIE: A Pipeline Energy-efficient Accelerator for Inference Process in Deep Neural Networks

Efficient B-mode Ultrasound Image Reconstruction Using CNN:Efficient B-mode Ultrasound Image Reconstruction from Sub-sampled RF Data using Deep Learning-matlab开发

Improving Nested Loop Pipelining on Coarse-Grained Reconfigurable Architectures

Area-Efficient Instruction Set Synthesis for Reconfigurable SoC Designs的翻译

Cognitive-Radio-Experiments-using-Reconfigurable-_Reconfigurable

Energy-Efficient Communication Protocol forWireless Microsensor Networks

FireFly A Reconfigurable Wireless Data Center Fabric

app-accelerator

A General-Purpose Reconfigurable Platform for Video and Image Processing

Design of Low-Power Coarse-Grained Reconfigurable Architectures

An On-Line Reconfigurable Four-Ary Tree-Based Network on Chip for Distributed Particle Filters

Orbital-angular-momentum-based reconfigurable optical switching and routing

All-optical reconfigurable multi-logic gates based on nonlinear polarization rotation effect in a single SOA

Impact of Crosstalk in an TFGB-OC-Based Reconfigurable Optical Add/Drop

On-chip reconfigurable mode converter based on cross-connected subwavelength Y-junctions

A Modeling and Mapping Method for Coarse/Fine Mixed-grained Reconfigurable Architecture

Modeling of Gaussian Network-Based Reconfigurable Network-on-Chip Designs

All-optically reconfigurable and tunable fiber surface grating for in-fiber devices: a wideband tunable filter

2018年密码芯片学术会议论文集

A polarization-agile aperture-coupled patch antenna using quad-mode reconfigurable feeding network