在PyTorch库的torch.nn模块中，提供了许多预定义的损失函数，用于不同的机器学习任务。以下是一些常见的损失函数：

MSELoss（均方误差损失）: 常用于回归问题，计算预测值与真实值之间的平均平方误差。

import torch.nn as nn  criterion = nn.MSELoss()  
output = ... # your network's output, shape should be the same as target  
loss = criterion(output, target)

CrossEntropyLoss（交叉熵损失）: 常用于分类问题，计算预测概率[logits]与真实标签之间的交叉熵。

import torch.nn as nn  criterion = nn.CrossEntropyLoss()  
output = ... # your network's output, shape should be (batch_size, num_classes)  
target = ... # target labels, shape should be (batch_size)  
loss = criterion(output, target)

BCELoss（二分类交叉熵损失）: 用于二分类问题，计算预测概率与真实标签之间的二进制交叉熵。

import torch.nn as nn  criterion = nn.BCELoss()  
output = ... # your network's output, shape should be (batch_size, 1)  
target = ... # target labels, shape should be (batch_size)  
loss = criterion(output, target)

NLLLoss（负对数似然损失）: 常用于多分类问题，计算预测概率与真实标签之间的负对数似然。

import torch.nn as nn  criterion = nn.NLLLoss()  
output = ... # your network's output, shape should be (batch_size, num_classes)  
target = ... # target labels, shape should be (batch_size)  
loss = criterion(output, target)

HingeEmbeddingLoss（铰链嵌入损失）: 用于铰链损失问题，计算预测值与真实值之间的铰链嵌入损失。

import torch.nn as nn  
criterion = nn.HingeEmbeddingLoss()  
output = ... # your network's output, shape should be the same as target  
target = ... # target labels, shape should be the same as output  
loss = criterion(output, target)

MultiMarginLoss（多边损失）: 常用于多分类问题，计算预测标签与真实标签之间的多边损失。

import torch.nn as nn  criterion = nn.MultiMarginLoss()  
output = ... # your network's output, shape should be (batch_size, num_classes)  
target = ... # target labels, shape should be (batch_size)  
loss = criterion(output, target)