昇騰AI處理器架構(gòu)與編程：深入理解CANN技術(shù)原理及應用

CONTENTS

\n

\n

目錄

\n

\n

第1章基礎理論00

\n

\n

1.1人工智能簡史00

\n

\n

1.2深度學習概論00

\n

\n

1.3神經(jīng)網(wǎng)絡理論0

\n

\n

1.3.1神經(jīng)元模型0

\n

\n

1.3.2感知機0

\n

\n

1.3.3多層感知機0

\n

\n

1.3.4卷積神經(jīng)網(wǎng)絡0

\n

\n

1.3.5應用示例0

\n

\n

第2章行業(yè)背景0

\n

\n

2.1神經(jīng)網(wǎng)絡芯片現(xiàn)狀0

\n

\n

2.1.1CPU0

\n

\n

2.1.2GPU0

\n

\n

2.1.3TPU0

\n

\n

2.1.4FPGA0

\n

\n

2.1.5昇騰AI處理器0

\n

\n

2.2神經(jīng)網(wǎng)絡芯片加速理論0

\n

\n

2.2.1GPU加速理論0

\n

\n

2.2.2TPU加速理論0

\n

\n

2.3深度學習框架0

\n

\n

2.3.1MindSpore0

\n

\n

2.3.2Caffe0

\n

\n

2.3.3TensorFlow0

\n

\n

2.3.4PyTorch0

\n

\n

2.4深度學習編譯框架——TVM0

\n

\n

\n

第3章硬件架構(gòu)0

\n

\n

3.1昇騰AI處理器總覽0

\n

\n

3.2達芬奇架構(gòu)0

\n

\n

3.2.1計算單元0

\n

\n

3.2.2存儲系統(tǒng)0

\n

\n

3.2.3控制單元0

\n

\n

3.2.4指令集設計0

\n

\n

3.3卷積加速原理0

\n

\n

3.3.1卷積加速0

\n

\n

3.3.2架構(gòu)對比0

\n

\n

\n

第4章軟件架構(gòu)0

\n

\n

4.1昇騰AI軟件棧總覽0

\n

\n

4.2神經(jīng)網(wǎng)絡軟件流

\n

\n

4.2.1流程編排器

\n

\n

4.2.2數(shù)字視覺預處理模塊

\n

\n

4.2.3張量加速引擎(TBE)

\n

\n

4.2.4運行管理器

\n

\n

4.2.5任務調(diào)度器

\n

\n

4.2.6框架管理器

\n

\n

4.2.7神經(jīng)網(wǎng)絡軟件流應用

\n

\n

4.3開發(fā)工具鏈

\n

\n

4.3.1功能簡介

\n

\n

4.3.2功能框架

\n

\n

4.3.3工具功能

\n

\n

第5章編程方法

\n

\n

5.1深度學習開發(fā)基礎

\n

\n

5.1.1深度學習編程理論

\n

\n

5.1.2深度學習推理優(yōu)化原理

\n

\n

5.1.3深度學習推理引擎

\n

\n

5.2昇騰AI軟件棧中的技術(shù)

\n

\n

5.2.1模型生成階段

\n

\n

5.2.2應用編譯與部署階段

\n

\n

5.3自定義算子開發(fā)

\n

\n

5.3.1開發(fā)步驟

\n

\n

5.3.2AI CPU算子開發(fā)

\n

\n

5.3.3AI Core算子開發(fā)

\n

\n

5.4自定義應用開發(fā) 

\n

\n

第6章實戰(zhàn)案例

\n

\n

6.1評價標準

\n

\n

6.1.1精度

\n

\n

6.1.2交并比

\n

\n

6.1.3均值平均精度

\n

\n

6.1.4吞吐量和時延

\n

\n

6.1.5能效比

\n

\n

6.2圖像識別

\n

\n

6.2.1數(shù)據(jù)集：ImageNet

\n

\n

6.2.2算法：ResNet

\n

\n

6.2.3模型遷移實踐

\n

\n

6.3目標檢測

\n

\n

6.3.1數(shù)據(jù)集：COCO

\n

\n

6.3.2算法：YoloV3

\n

\n

6.3.3自定義算子實踐

\n

\n

附錄A縮略詞列表

\n

\n

附錄BAscend開發(fā)者社區(qū)及資料下載

\n

\n

附錄C智能開發(fā)平臺ModelArts簡介

\n

\n