fix utf8 (apache#6045)

src/operator/lrn.cc

@@ -42,7 +42,7 @@ MXNET_REGISTER_OP_PROPERTY(LRN, LocalResponseNormProp)
 .add_arguments(LRNParam::__FIELDS__())
 .describe(R"code(Applies local response normalization to the input.
 
-The local response normalization layer performs “lateral inhibition” by normalizing 
+The local response normalization layer performs "lateral inhibition" by normalizing 
 over local input regions. 
 
 If :math:`a_{x,y}^{i}` is the activity of a neuron computed by applying kernel :math:`i` at position
@@ -52,7 +52,7 @@ activity :math:`b_{x,y}^{i}` is given by the expression:
 .. math::   
    b_{x,y}^{i} = \frac{a_{x,y}^{i}}{\Bigg({k + \alpha \sum_{j=max(0, i-\frac{n}{2})}^{min(N-1, i+\frac{n}{2})} (a_{x,y}^{j})^{2}}\Bigg)^{\beta}}
 
-where the sum runs over :math:`n` “adjacent” kernel maps at the same spatial position, and :math:`N` is the total
+where the sum runs over :math:`n` "adjacent" kernel maps at the same spatial position, and :math:`N` is the total
 number of kernels in the layer.
 
 )code" ADD_FILELINE);

src/operator/roi_pooling.cc

@@ -245,7 +245,7 @@ The cropped feature maps are pooled by standard max pooling operation to a fixed
 indicated by a `pooled_size` parameter. batch_size will change to the number of region
 bounding boxes after `ROIPooling`.
 
-The size of each region of interest doesn’t have to be perfectly divisible by
+The size of each region of interest doesn't have to be perfectly divisible by
 the number of pooling sections(`pooled_size`).
 
 Example::

src/operator/tensor/matrix_op.cc

@@ -515,7 +515,7 @@ there cases:
                            [ 1.,  2.,  1.,  2.,  1.,  2.],
                            [ 3.,  4.,  3.,  4.,  3.,  4.]]
 
-- **n>d**. ``reps`` is promoted to length *n* by pre-pending 1’s to it. Thus for
+- **n>d**. ``reps`` is promoted to length *n* by pre-pending 1's to it. Thus for
   an input shape ``(2,3)``, ``repos=(2,)`` is treated as ``(1,2)``::