相似性最大化简介

1import torch
2import torch.nn as nn
3import matplotlib.pyplot as plt
4
5model = nn.Sequential(nn.Linear(5,3), # 模型定义
6                      nn.ReLU(),
7                      nn.Linear(3, 2),
8                      nn.ReLU())
9
10x = torch.FloatTensor([[1, 0, 0, 1, 1], # 人工输入数据
11                       [0, 0, 1, 1, 1],
12                       [1, 1, 1, 1, 1],
13                       [1, 0, 1, 0, 1],
14                       [1, 0, 0, 0, 1],
15                       [0, 0, 1, 0, 0]])
16
17t1_x = x + 0.3*torch.randn(6, 5) # 增强版本 1
18t2_x =  x + 0.3*torch.randn(6, 5) # 增强版本 2
19
20optimizer = torch.optim.Adam(model.parameters()) # 优化器
21criterion = nn.CosineSimilarity(dim=-1) # 余弦相似度目标函数
22
23for epoch in range(10):
24  optimizer.zero_grad()
25  f1, f2 = model(t1_x), model(t2_x) # 提取特征
26  loss = -criterion(f1, f2).mean() # 最大化余弦相似度
27  loss.backward()
28  optimizer.step() # 更新模型参数
29
30  with torch.no_grad():
31    f = model(x)
32    mean = f.mean(0)
33    var = f.var(0)
34    print("第 {} 轮特征均值\n".format(epoch), mean.numpy())
35    print("第 {} 轮特征方差\n".format(epoch), var.numpy())
36    print("=====================")