update autodiff.

NailinLiao · Oct 2, 2022 · 7c221b3 · 7c221b3
1 parent b209173
commit 7c221b3
Show file tree

Hide file tree

Showing 25 changed files with 10 additions and 6 deletions.
diff --git a/Frontend/AutoDiff/05.forward_mode.ipynb b/Frontend/AutoDiff/05.forward_mode.ipynb
@@ -19,9 +19,11 @@
     "\n",
     "> 前向模式（Forward Automatic Differentiation，也叫做 tangent mode AD）或者前向累积梯度（前向模式）\n",
     "\n",
+    "![](images/[email protected])\n",
+    "\n",
     "前向自动微分中，从计算图的起点开始，沿着计算图边的方向依次向前计算，最终到达计算图的终点。它根据自变量的值计算出计算图中每个节点的值 以及其导数值，并保留中间结果。一直得到整个函数的值和其导数值。整个过程对应于一元复合函数求导时从最内层逐步向外层求导。\n",
     "\n",
-    "![](images/WX20220522-180607@2x.png)\n",
+    "![](images/forward_mode.png)\n",
     "\n",
     "简单确实简单，可以总结前向自动微分关键步骤为：\n",
     "\n",
@@ -123,7 +125,7 @@
     "            dx = self.dx - other.dx\n",
     "        elif isinstance(other, float):\n",
     "            x = self.x - other\n",
-    "            ex = self.dx\n",
+    "            dx = self.dx\n",
     "        else:\n",
     "            return NotImplementedError\n",
     "        return ADTangent(x, dx)\n",
@@ -154,7 +156,7 @@
    "cell_type": "markdown",
    "metadata": {},
    "source": [
-    "以公式5为例：\n",
+    "以下面为例：\n",
     "\n",
     "$$\n",
     "f(x1,x2)=ln(x1)+x1x2−sin(x2) \\tag{1}\n",
@@ -179,9 +181,9 @@
     }
    ],
    "source": [
-    "x = ADTangent(x=2., dx=1)\n",
-    "y = ADTangent(x=5., dx=0)\n",
-    "f = ADTangent.log(x) + x * y - ADTangent.sin(y)\n",
+    "x1 = ADTangent(x=2., dx=1)\n",
+    "x2 = ADTangent(x=5., dx=0)\n",
+    "f = ADTangent.log(x) + x * x2 - ADTangent.sin(x2)\n",
     "print(f)"
    ]
   },

diff --git a/Frontend/AutoDiff/images/.DS_Store b/Frontend/AutoDiff/images/.DS_Store
diff --git a/Frontend/AutoDiff/images/[email protected] b/Frontend/AutoDiff/images/[email protected]
diff --git a/Frontend/AutoDiff/images/[email protected] b/Frontend/AutoDiff/images/[email protected]
diff --git a/Frontend/AutoDiff/images/[email protected] b/Frontend/AutoDiff/images/[email protected]
diff --git a/Frontend/AutoDiff/images/[email protected] b/Frontend/AutoDiff/images/[email protected]
diff --git a/Frontend/AutoDiff/images/[email protected] b/Frontend/AutoDiff/images/[email protected]
diff --git a/Frontend/AutoDiff/images/[email protected] b/Frontend/AutoDiff/images/[email protected]
diff --git a/Frontend/AutoDiff/images/ast.graffle b/Frontend/AutoDiff/images/ast.graffle
diff --git a/Frontend/AutoDiff/images/ast.png b/Frontend/AutoDiff/images/ast.png
diff --git a/Frontend/AutoDiff/images/forward mode.graffle b/Frontend/AutoDiff/images/forward mode.graffle
diff --git a/Frontend/AutoDiff/images/forward_mode.png b/Frontend/AutoDiff/images/forward_mode.png
diff --git a/Frontend/AutoDiff/images/jacobian.graffle b/Frontend/AutoDiff/images/jacobian.graffle
diff --git a/Frontend/AutoDiff/images/jacobian.png b/Frontend/AutoDiff/images/jacobian.png
diff --git a/Frontend/AutoDiff/images/operator_overload.graffle b/Frontend/AutoDiff/images/operator_overload.graffle
diff --git a/Frontend/AutoDiff/images/operator_overload.png b/Frontend/AutoDiff/images/operator_overload.png
diff --git a/Frontend/AutoDiff/images/parameter.graffle b/Frontend/AutoDiff/images/parameter.graffle
diff --git a/Frontend/AutoDiff/images/parameter.png b/Frontend/AutoDiff/images/parameter.png
diff --git a/Frontend/AutoDiff/images/reversed mode.graffle b/Frontend/AutoDiff/images/reversed mode.graffle
diff --git a/Frontend/AutoDiff/images/reversed_mode.png b/Frontend/AutoDiff/images/reversed_mode.png
diff --git a/Frontend/AutoDiff/images/tangent_adjoint.graffle b/Frontend/AutoDiff/images/tangent_adjoint.graffle
diff --git a/Frontend/AutoDiff/images/tangent_adjoint.png b/Frontend/AutoDiff/images/tangent_adjoint.png
diff --git a/Frontend/Foundation/02.fundamentals.pptx b/Frontend/Foundation/02.fundamentals.pptx
diff --git a/Frontend/Foundation/03.history.pptx b/Frontend/Foundation/03.history.pptx
diff --git a/README.md b/README.md
@@ -30,6 +30,8 @@
 |1|自动微分|01 基本介绍|[silde](./Frontend/AutoDiff/01.introduction.pptx), [video](https://www.bilibili.com/video/BV1FV4y1T7zp/), [article](https://zhuanlan.zhihu.com/p/518198564)|
 | |自动微分|02 什么是微分|[silde](./Frontend/AutoDiff/02.base_concept.pptx), [video](https://www.bilibili.com/video/BV1Ld4y1M7GJ/), [article](https://zhuanlan.zhihu.com/p/518198564)|
 | |自动微分|03 数学计算模式|[silde](./Frontend/AutoDiff/03.grad_mode.pptx), [video](https://www.bilibili.com/video/BV1zD4y117bL/), [article](https://zhuanlan.zhihu.com/p/518296942)|
+| |自动微分|04 三种实现方式|[silde](./Frontend/AutoDiff/04.grad_mode.pptx), [video](https://www.bilibili.com/video/BV1BN4y1P76t/), [article](https://zhuanlan.zhihu.com/p/520065656)|
+| |自动微分|05 手把手实现正向微分框架|[silde](./Frontend/AutoDiff/05.forward_mode.ipynb), [video](https://www.bilibili.com/video/BV1Ne4y1p7WU/), [article](https://zhuanlan.zhihu.com/p/520451681)|
 |||||
 
 **底层编译技术**