notebook updates

machine-learning-hats/notebooks/4-conv2d.ipynb

@@ -55,14 +55,13 @@
     "### Convolution Operation\n",
     "Two-dimensional convolutional layer for image height $H$, width $W$, number of input channels $C$, number of output kernels (filters) $N$, and kernel height $J$ and width $K$ is given by:\n",
     "\n",
-    "\\begin{align}\n",
-    "\\label{convLayer}\n",
-    "\\boldsymbol{Y}[v,u,n] &= \\boldsymbol{\\beta}[n] + \\sum_{c=1}^{C} \\sum_{j=1}^{J} \\sum_{k=1}^{K} \\boldsymbol{X}[v+j,u+k,c]\\, \\boldsymbol{W}[j,k,c,n]\\,,\n",
-    "\\end{align}\n",
+    "$$\n",
+    "\\boldsymbol{Y}[v,u,n] = \\boldsymbol{\\beta}[n] + \\sum_{c=1}^{C} \\sum_{j=1}^{J} \\sum_{k=1}^{K} \\boldsymbol{X}[v+j,u+k,c]\\, \\boldsymbol{W}[j,k,c,n]\\,,\n",
+    "$$\n",
     "\n",
     "where $Y$ is the output tensor of size $V \\times U \\times N$, $W$ is the weight tensor of size $J \\times K \\times C \\times N$ and $\\beta$ is the bias vector of length $N$ .\n",
     "\n",
-    "The example below has $C=1$ input channel and $N=1$ ($J\\times K=3\\times 3$) kernel [credit](https://towardsdatascience.com/types-of-convolution-kernels-simplified-f040cb307c37):\n",
+    "The example below has $C=1$ input channel and $N=1$ ($J\\times K=3\\times 3$) kernel ([credit](https://towardsdatascience.com/types-of-convolution-kernels-simplified-f040cb307c37)):\n",
     "\n",
     "![convolution](https://miro.medium.com/v2/resize:fit:780/1*Eai425FYQQSNOaahTXqtgg.gif)"
    ]
@@ -84,7 +83,7 @@
    "source": [
     "### Pooling\n",
     "\n",
-    "We also add pooling layers to reduce the image size between layers. For example, max pooling: (also from [here]([page](https://cs231n.github.io/convolutional-networks/))\n",
+    "We also add pooling layers to reduce the image size between layers. For example, max pooling (also from [here]([page](https://cs231n.github.io/convolutional-networks/))):\n",
     "\n",
     "![maxpool](https://cs231n.github.io/assets/cnn/maxpool.jpeg)"
    ]