diff --git a/unsuper/Readme.md b/unsuper/Readme.md
index ff7f156..d614bde 100644
--- a/unsuper/Readme.md
+++ b/unsuper/Readme.md
@@ -1,9 +1,35 @@
-1. 3x3的时候会有重复
-  1. 重复的权重，虽然权重看起来都一样，但是有稍微的不同，不是完全一样
-  2. 3x3太小了导致了样本的信噪比太低，大部分的样本切出来都是0
-2. 5x5的时候会有网格状重复
-3. 7x7的时候边框区域问题
 
 1. 输入的信噪比
 2. loss函数的设计
-3. grad信息的应用
\ No newline at end of file
+3. grad信息的应用
+
+
+abs.cpu().detach().numpy()
+array([[8.1206687e-02, 2.2388995e-05, 3.7080176e-02, 5.7033218e-02,
+        1.7404296e-03, 7.6270252e-02, 5.9453689e-02, 4.0801242e-05]],
+      dtype=float32)
+ratio_nor.cpu().detach().numpy()
+array([[4.3121886e+00, 3.2778070e-07, 8.9907879e-01, 2.1270120e+00,
+        1.9807382e-03, 3.8038602e+00, 2.3113825e+00, 1.0885816e-06]],
+      dtype=float32)
+
+都比较差的时候区分不开 
+
+
+## 发现的问题
+1. 3x3的时候会有重复
+   1. 重复的权重，虽然权重看起来都一样，但是有稍微的不同，不是完全一样
+   2. 3x3太小了导致了样本的信噪比太低，大部分的样本切出来都是0
+2. 5x5的时候会有网格状重复
+3. 7x7的时候边框区域问题
+
+## 发现的原因
+1. 几个卷积核都是重复的，网格状的
+   1. grad太大，learning rate 太大
+   2. grad太小，训练的epoch不合适
+
+## 可能的策略
+1. 每个卷积核的改变权重(grad)能量守恒
+2. 卷积核的每个像素的权重都独立统计？
+   1. 权重的reduce体现的是相互之间的可比性，关系
+3. 需要考虑梯度的绝对比值？
\ No newline at end of file
diff --git a/unsuper/minist.py b/unsuper/minist.py
index 33e9158..78eb13b 100644
--- a/unsuper/minist.py
+++ b/unsuper/minist.py
@@ -20,8 +20,8 @@ np.random.seed(seed)
 random.seed(seed)
 
 
-# device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
-device = torch.device("cpu")
+device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
+# device = torch.device("cpu")
 # device = torch.device("mps")
 
 num_epochs = 1
@@ -110,11 +110,27 @@ model = ConvNet().to(device)
 model.train()
 
 # Train the model unsuper
-epochs = 2
+epochs = 3
 n_total_steps = len(train_loader)
 for epoch in range(epochs):
     for i, (images, labels) in enumerate(train_loader):
         images = images.to(device)
+
+        # images = torch.ones((1, 1, 5, 5), device=device)
+        # type = random.randint(0, 10)
+        # if type == 0:
+        #     rand = random.randint(0, 4)
+        #     images[:, :, rand, :] = images[:, :, rand, :] * 0.5
+        # if type == 1:
+        #     rand = random.randint(0, 4)
+        #     images[:, :, :, rand] = images[:, :, :, rand] * 0.5
+        # if type == 2:
+        #     images[:, :, 0, 0] = images[:, :, 0, 0] * 0.5
+        #     images[:, :, 1, 1] = images[:, :, 1, 1] * 0.5
+        #     images[:, :, 2, 2] = images[:, :, 2, 2] * 0.5
+        #     images[:, :, 3, 3] = images[:, :, 3, 3] * 0.5
+        #     images[:, :, 4, 4] = images[:, :, 4, 4] * 0.5
+
         outputs = model.forward_unsuper(images)
 
         outputs = outputs.permute(0, 2, 3, 1)  # 64 8 24 24 -> 64 24 24 8
@@ -137,7 +153,9 @@ for epoch in range(epochs):
         model.conv1.weight.grad = None
         loss.backward()
 
-        model.conv1.weight.data = model.conv1.weight.data - model.conv1.weight.grad * 1000
+        # if epoch >= (epochs - 1):
+        #     continue
+        model.conv1.weight.data = model.conv1.weight.data - model.conv1.weight.grad * 0.001
         model.conv1.weight.data = model.normal_conv1_weight()
 
         if (i + 1) % 100 == 0: