# Differences with torch.nn.LocalResponseNorm

torch.nn.LocalResponseNorm

class torch.nn.LocalResponseNorm(
    size,
    alpha=0.0001,
    beta=0.75,
    k=1.0
)(input) -> Tensor

For more information, see torch.nn.LocalResponseNorm.

mindspore.nn.LRN

class mindspore.nn.LRN(
    depth_radius=5,
    bias=1.0,
    alpha=1.0,
    beta=0.5,
    norm_region="ACROSS_CHANNELS"
)(x) -> Tensor

For more information, see mindspore.nn.LRN.

Differences

PyTorch: This API performs Local Response Normalization (LRN) operation that normalizes the input for each neuron in a specific way to improve the generalization ability of deep neural networks. It returns a tensor with the same type as the input.

MindSpore: It implements the same functionality as PyTorch, but with different parameter names. The depth_radius parameter in MindSpore performs the same function as the size parameter in PyTorch, and there is a mapping relationship of twice the value: size=2*depth_radius. Currently, mindspore.nn.LRN and tf.raw_ops.LRN can be completely aligned, and both can achieve the same accuracy. However, if compared with torch.nn.LocalResponseNorm, there may be a precision difference of 1e-3.

Categories	Subcategories	PyTorch	MindSpore	Difference
Parameters	Parameters 1	size	depth_radius	The number of adjacent neurons to consider for normalization,mapping relationship: size=2*depth_radius
	Parameter 2	k	bias	Same function, different parameter names
	Parameter 3	alpha	alpha	-
	Parameter 4	beta	beta	-
	Parameter 5	-	norm_region	Specify the norm region, PyTorch doesn’t have this parameter
Input	Single input	input	x	Same function, different parameter names

Code Example 1

Thedepth_radius in MindSpore corresponds to size in PyTorch with its value halved. It is necessary to set depth_radius as half of size to achieve the same function.

# PyTorch
import torch
import numpy as np

input_x = torch.from_numpy(np.array([[[[2.4], [3.51]],[[1.3], [-4.4]]]], dtype=np.float32))
output = torch.nn.LocalResponseNorm(size=2, alpha=0.0001, beta=0.75, k=1.0)(input_x)
print(output.numpy())
#[[[[ 2.3994818]
#   [ 3.5083795]]
#  [[ 1.2996368]
#   [-4.39478  ]]]]

# MindSpore
import mindspore
from mindspore import Tensor
import numpy as np

input_x = Tensor(np.array([[[[2.4], [3.51]],[[1.3], [-4.4]]]]), mindspore.float32)
lrn = mindspore.nn.LRN(depth_radius=1, bias=1.0, alpha=0.0001, beta=0.75)
output = lrn(input_x)
print(output)
#[[[[ 2.39866  ]
#   [ 3.5016835]]
#  [[ 1.2992741]
#   [-4.3895745]]]]