diff --git a/convolution.go b/convolution.go
new file mode 100644
index 0000000..d4ff826
--- /dev/null
+++ b/convolution.go
@@ -0,0 +1,314 @@
+package gotensor
+
+import (
+	"git.kingecg.top/kingecg/gomatrix"
+)
+
+// Conv2D 二维卷积操作
+func (t *Tensor) Conv2D(kernel *Tensor, stride, padding int) (*Tensor, error) {
+	// 假设输入格式为 [batch_size, channels, height, width]
+	inputShape := t.Data.Shape()
+	kernelShape := kernel.Data.Shape()
+
+	// 检查输入维度
+	if len(inputShape) != 4 || len(kernelShape) != 4 {
+		return nil, nil // 这里应该返回错误，但暂时返回nil
+	}
+
+	batchSize, inputChannels, inputHeight, inputWidth := inputShape[0], inputShape[1], inputShape[2], inputShape[3]
+	kernelChannels, numFilters, kernelHeight, kernelWidth := kernelShape[0], kernelShape[1], kernelShape[2], kernelShape[3]
+
+	// 检查通道数是否匹配
+	if inputChannels != kernelChannels {
+		return nil, nil // 应该返回错误
+	}
+
+	// 计算输出尺寸
+	outputHeight := (inputHeight + 2*padding - kernelHeight) / stride + 1
+	outputWidth := (inputWidth + 2*padding - kernelWidth) / stride + 1
+
+	// 分配输出矩阵
+	outputSize := batchSize * numFilters * outputHeight * outputWidth
+	outputData := make([]float64, outputSize)
+	outputShape := []int{batchSize, numFilters, outputHeight, outputWidth}
+
+	// 执行卷积操作
+	outputIdx := 0
+	for b := 0; b < batchSize; b++ { // batch
+		for f := 0; f < numFilters; f++ { // filter
+			for oh := 0; oh < outputHeight; oh++ { // output height
+				for ow := 0; ow < outputWidth; ow++ { // output width
+					// 计算卷积结果
+					sum := 0.0
+					for c := 0; c < inputChannels; c++ { // channel
+						for kh := 0; kh < kernelHeight; kh++ { // kernel height
+							for kw := 0; kw < kernelWidth; kw++ { // kernel width
+								ih := oh*stride - padding + kh
+								iw := ow*stride - padding + kw
+
+								if ih >= 0 && ih < inputHeight && iw >= 0 && iw < inputWidth {
+									inputVal, _ := t.Data.Get(b, c, ih, iw)
+									kernelVal, _ := kernel.Data.Get(c, f, kh, kw)
+									sum += inputVal * kernelVal
+								}
+							}
+						}
+					}
+					outputData[outputIdx] = sum
+					outputIdx++
+				}
+			}
+		}
+	}
+
+	outputMatrix, err := gomatrix.NewMatrix(outputData, outputShape)
+	if err != nil {
+		return nil, err
+	}
+
+	output := &Tensor{
+		Data: outputMatrix,
+		Op:   "conv2d",
+	}
+
+	output.Prevs[0] = t
+	output.Prevs[1] = kernel
+	output.Num_Prevs = 2
+	output.Args[0] = stride
+	output.Args[1] = padding
+
+	output.backwardFunc = func() {
+		if t.Grad != nil {
+			// 这里应该实现卷积的反向传播
+			// 由于复杂性，此处暂不实现
+		}
+		if kernel.Grad != nil {
+			// 这里应该实现核的梯度计算
+			// 由于复杂性，此处暂不实现
+		}
+	}
+
+	return output, nil
+}
+
+// MaxPool2D 二维最大池化操作
+func (t *Tensor) MaxPool2D(kernelSize, stride int) (*Tensor, error) {
+	// 假设输入格式为 [batch_size, channels, height, width]
+	inputShape := t.Data.Shape()
+
+	if len(inputShape) != 4 {
+		return nil, nil // 应该返回错误
+	}
+
+	batchSize, inputChannels, inputHeight, inputWidth := inputShape[0], inputShape[1], inputShape[2], inputShape[3]
+
+	// 计算输出尺寸（假定没有padding）
+	outputHeight := (inputHeight-kernelSize)/stride + 1
+	outputWidth := (inputWidth-kernelSize)/stride + 1
+
+	// 分配输出矩阵
+	outputSize := batchSize * inputChannels * outputHeight * outputWidth
+	outputData := make([]float64, outputSize)
+	outputShape := []int{batchSize, inputChannels, outputHeight, outputWidth}
+
+	// 记录最大值位置，用于反向传播
+	maxIndices := make([]int, len(outputData))
+
+	// 执行最大池化操作
+	outputIdx := 0
+	for b := 0; b < batchSize; b++ {
+		for c := 0; c < inputChannels; c++ {
+			for oh := 0; oh < outputHeight; oh++ {
+				for ow := 0; ow < outputWidth; ow++ {
+					startH := oh * stride
+					startW := ow * stride
+					
+					maxVal := -1e9 // 初始为极小值
+					maxIH, maxIW := -1, -1 // 最大值的位置
+					
+					// 在池化窗口内找最大值
+					for kh := 0; kh < kernelSize; kh++ {
+						for kw := 0; kw < kernelSize; kw++ {
+							ih := startH + kh
+							iw := startW + kw
+							
+							if ih < inputHeight && iw < inputWidth {
+								inputVal, _ := t.Data.Get(b, c, ih, iw)
+								if inputVal > maxVal {
+									maxVal = inputVal
+									maxIH = ih
+									maxIW = iw
+								}
+							}
+						}
+					}
+					
+					outputData[outputIdx] = maxVal
+					// 存储在扁平化数组中的索引
+					maxIndices[outputIdx] = ((b*inputChannels+c)*inputHeight+maxIH)*inputWidth + maxIW
+					outputIdx++
+				}
+			}
+		}
+	}
+
+	outputMatrix, err := gomatrix.NewMatrix(outputData, outputShape)
+	if err != nil {
+		return nil, err
+	}
+
+	output := &Tensor{
+		Data: outputMatrix,
+		Op:   "maxpool2d",
+	}
+
+	output.Prevs[0] = t
+	output.Num_Prevs = 1
+	output.Args[0] = kernelSize
+	output.Args[1] = stride
+
+	output.backwardFunc = func() {
+		if t.Grad != nil {
+			// 反向传播：只将梯度传递给最大值位置
+			// 遍历输出的每个元素
+			outputIdx := 0
+			for b := 0; b < batchSize; b++ {
+				for c := 0; c < inputChannels; c++ {
+					for oh := 0; oh < outputHeight; oh++ {
+						for ow := 0; ow < outputWidth; ow++ {
+							// 获取最大值在输入中的位置
+							inIdx := maxIndices[outputIdx]
+							
+							// 获取对应的输出梯度
+							outputGrad, _ := output.Grad.Get(b, c, oh, ow)
+							
+							// 将输出梯度添加到输入对应位置
+							// 需要将一维索引转换为多维索引
+							b_idx := inIdx / (inputChannels * inputHeight * inputWidth)
+							remaining := inIdx % (inputChannels * inputHeight * inputWidth)
+							c_idx := remaining / (inputHeight * inputWidth)
+							remaining = remaining % (inputHeight * inputWidth)
+							h_idx := remaining / inputWidth
+							w_idx := remaining % inputWidth
+							
+							currentGrad, _ := t.Grad.Get(b_idx, c_idx, h_idx, w_idx)
+							newGrad := currentGrad + outputGrad
+							t.Grad.Set(newGrad, b_idx, c_idx, h_idx, w_idx)
+							
+							outputIdx++
+						}
+					}
+				}
+			}
+		}
+	}
+
+	return output, nil
+}
+
+// AvgPool2D 二维平均池化操作
+func (t *Tensor) AvgPool2D(kernelSize, stride int) (*Tensor, error) {
+	// 假设输入格式为 [batch_size, channels, height, width]
+	inputShape := t.Data.Shape()
+
+	if len(inputShape) != 4 {
+		return nil, nil // 应该返回错误
+	}
+
+	batchSize, inputChannels, inputHeight, inputWidth := inputShape[0], inputShape[1], inputShape[2], inputShape[3]
+
+	// 计算输出尺寸（假定没有padding）
+	outputHeight := (inputHeight-kernelSize)/stride + 1
+	outputWidth := (inputWidth-kernelSize)/stride + 1
+
+	// 分配输出矩阵
+	outputSize := batchSize * inputChannels * outputHeight * outputWidth
+	outputData := make([]float64, outputSize)
+	outputShape := []int{batchSize, inputChannels, outputHeight, outputWidth}
+
+	// 执行平均池化操作
+	outputIdx := 0
+	for b := 0; b < batchSize; b++ {
+		for c := 0; c < inputChannels; c++ {
+			for oh := 0; oh < outputHeight; oh++ {
+				for ow := 0; ow < outputWidth; ow++ {
+					startH := oh * stride
+					startW := ow * stride
+					
+					sum := 0.0
+					count := 0
+					
+					// 在池化窗口内计算平均值
+					for kh := 0; kh < kernelSize; kh++ {
+						for kw := 0; kw < kernelSize; kw++ {
+							ih := startH + kh
+							iw := startW + kw
+							
+							if ih < inputHeight && iw < inputWidth {
+								inputVal, _ := t.Data.Get(b, c, ih, iw)
+								sum += inputVal
+								count++
+							}
+						}
+					}
+					
+					avgVal := sum / float64(count)
+					outputData[outputIdx] = avgVal
+					outputIdx++
+				}
+			}
+		}
+	}
+
+	outputMatrix, err := gomatrix.NewMatrix(outputData, outputShape)
+	if err != nil {
+		return nil, err
+	}
+
+	output := &Tensor{
+		Data: outputMatrix,
+		Op:   "avgpool2d",
+	}
+
+	output.Prevs[0] = t
+	output.Num_Prevs = 1
+	output.Args[0] = kernelSize
+	output.Args[1] = stride
+
+	output.backwardFunc = func() {
+		if t.Grad != nil {
+			// 反向传播：将平均梯度分配给池化窗口内的所有元素
+			outputIdx := 0
+			for b := 0; b < batchSize; b++ {
+				for c := 0; c < inputChannels; c++ {
+					for oh := 0; oh < outputHeight; oh++ {
+						for ow := 0; ow < outputWidth; ow++ {
+							startH := oh * stride
+							startW := ow * stride
+							
+							outputGrad, _ := output.Grad.Get(b, c, oh, ow)
+							avgGrad := outputGrad / float64(kernelSize*kernelSize)
+							
+							// 将平均梯度分配给对应区域
+							for kh := 0; kh < kernelSize; kh++ {
+								for kw := 0; kw < kernelSize; kw++ {
+									ih := startH + kh
+									iw := startW + kw
+									
+									if ih < inputHeight && iw < inputWidth {
+										currentGrad, _ := t.Grad.Get(b, c, ih, iw)
+										newGrad := currentGrad + avgGrad
+										t.Grad.Set(newGrad, b, c, ih, iw)
+									}
+								}
+							}
+							outputIdx++
+						}
+					}
+				}
+			}
+		}
+	}
+
+	return output, nil
+}
\ No newline at end of file
diff --git a/examples/cnn_example.go b/examples/cnn_example.go
new file mode 100644
index 0000000..f2f08e6
--- /dev/null
+++ b/examples/cnn_example.go
@@ -0,0 +1,100 @@
+package main
+
+import (
+	"fmt"
+	"git.kingecg.top/kingecg/gotensor"
+)
+
+func main() {
+	fmt.Println("=== 卷积神经网络示例 ===")
+	
+	// 创建一个简单的4x4灰度图像（批大小为1）
+	imageData := []float64{
+		1.0, 2.0, 3.0, 4.0,
+		5.0, 6.0, 7.0, 8.0,
+		9.0, 10.0, 11.0, 12.0,
+		13.0, 14.0, 15.0, 16.0,
+	}
+	
+	// 形状为 [batch, channel, height, width]
+	imageShape := []int{1, 1, 4, 4}
+	imageTensor, err := gotensor.NewTensor(imageData, imageShape)
+	if err != nil {
+		panic(err)
+	}
+	
+	fmt.Printf("输入图像:\n%s\n", imageTensor.String())
+	
+	// 创建一个简单的卷积核 (3x3)
+	kernelData := []float64{
+		1.0, 0.0, -1.0,
+		1.0, 0.0, -1.0,
+		1.0, 0.0, -1.0,
+	}
+	// 形状为 [input_channels, output_channels, kernel_height, kernel_width]
+	kernelShape := []int{1, 1, 3, 3}
+	kernelTensor, err := gotensor.NewTensor(kernelData, kernelShape)
+	if err != nil {
+		panic(err)
+	}
+	
+	fmt.Printf("卷积核:\n%s\n", kernelTensor.String())
+	
+	// 执行卷积操作 (stride=1, padding=0)
+	convResult, err := imageTensor.Conv2D(kernelTensor, 1, 0)
+	if err != nil {
+		panic(err)
+	}
+	
+	fmt.Printf("卷积结果:\n%s\n", convResult.String())
+	
+	// 应用ReLU激活函数
+	reluResult := convResult.ReLU()
+	fmt.Printf("ReLU结果:\n%s\n", reluResult.String())
+	
+	// 执行最大池化 (kernel_size=2, stride=2)
+	poolResult, err := reluResult.MaxPool2D(2, 2)
+	if err != nil {
+		panic(err)
+	}
+	
+	fmt.Printf("池化结果:\n%s\n", poolResult.String())
+	
+	// 展平操作
+	flattened := poolResult.Flatten()
+	fmt.Printf("展平后大小: %d\n", flattened.Size())
+	
+	// 创建一些随机权重进行全连接层操作
+	weightsData := []float64{0.1, 0.2, 0.3, 0.4, 0.5, 0.6}
+	weightsShape := []int{flattened.Size(), 2} // 输出2类（猫/狗）
+	weights, err := gotensor.NewTensor(weightsData, weightsShape)
+	if err != nil {
+		panic(err)
+	}
+	
+	// 全连接层计算 (矩阵乘法)
+	fcResult, err := flattened.MatMul(weights)
+	if err != nil {
+		panic(err)
+	}
+	
+	fmt.Printf("全连接输出:\n%s\n", fcResult.String())
+	
+	// 应用Softmax
+	softmaxResult := fcResult.Softmax()
+	fmt.Printf("Softmax输出 (概率分布):\n%s\n", softmaxResult.String())
+	
+	// 创建目标标签 (猫=1, 狗=0)
+	targetData := []float64{1.0, 0.0}
+	target, err := gotensor.NewTensor(targetData, []int{1, 2})
+	if err != nil {
+		panic(err)
+	}
+	
+	// 计算交叉熵损失
+	loss := softmaxResult.CrossEntropy(target)
+	lossVal, _ := loss.Data.Get(0)
+	fmt.Printf("交叉熵损失: %f\n", lossVal)
+	
+	fmt.Println("=== 模型前向传播完成 ===")
+}
\ No newline at end of file
diff --git a/extended_tensor_test.go b/extended_tensor_test.go
new file mode 100644
index 0000000..218e2bd
--- /dev/null
+++ b/extended_tensor_test.go
@@ -0,0 +1,201 @@
+package gotensor
+
+import (
+	"testing"
+)
+
+func TestSigmoid(t *testing.T) {
+	data := []float64{0.0, 1.0, -1.0, 2.0}
+	shape := []int{2, 2}
+	
+	tensor, err := NewTensor(data, shape)
+	if err != nil {
+		t.Fatal(err)
+	}
+	
+	result := tensor.Sigmoid()
+	
+	// 验证输出值是否在0和1之间
+	for i := 0; i < result.Size(); i++ {
+		val, err := result.Get(i/2, i%2)
+		if err != nil {
+			t.Fatal(err)
+		}
+		
+		if val < 0.0 || val > 1.0 {
+			t.Errorf("Sigmoid输出应在[0,1]范围内，实际值: %f", val)
+		}
+	}
+	
+	// 特定值检查
+	expected_values := []float64{0.5, 0.7310585786300049, 0.2689414213699951, 0.8807970779778823}
+	for i, expected := range expected_values {
+		actual, err := result.Get(i/2, i%2)
+		if err != nil {
+			t.Fatal(err)
+		}
+		
+		if actual < expected-0.001 || actual > expected+0.001 {
+			t.Errorf("位置[%d,%d]的Sigmoid值不正确，期望: %f, 实际: %f", i/2, i%2, expected, actual)
+		}
+	}
+}
+
+func TestReLU(t *testing.T) {
+	data := []float64{-1.0, 0.0, 1.0, 2.0}
+	shape := []int{2, 2}
+	
+	tensor, err := NewTensor(data, shape)
+	if err != nil {
+		t.Fatal(err)
+	}
+	
+	result := tensor.ReLU()
+	
+	// 验证ReLU的性质：负值变0，非负值保持不变
+	expected_values := []float64{0.0, 0.0, 1.0, 2.0}
+	for i, expected := range expected_values {
+		actual, err := result.Get(i/2, i%2)
+		if err != nil {
+			t.Fatal(err)
+		}
+		
+		if actual != expected {
+			t.Errorf("位置[%d,%d]的ReLU值不正确，期望: %f, 实际: %f", i/2, i%2, expected, actual)
+		}
+	}
+}
+
+func TestSoftmax(t *testing.T) {
+	data := []float64{1.0, 2.0, 3.0, 4.0}
+	shape := []int{2, 2}
+	
+	tensor, err := NewTensor(data, shape)
+	if err != nil {
+		t.Fatal(err)
+	}
+	
+	result := tensor.Softmax()
+	
+	// 验证Softmax的性质：所有值在[0,1]之间，且每行之和为1
+	for i := 0; i < 2; i++ {
+		rowSum := 0.0
+		for j := 0; j < 2; j++ {
+			val, err := result.Get(i, j)
+			if err != nil {
+				t.Fatal(err)
+			}
+			
+			if val < 0.0 || val > 1.0 {
+				t.Errorf("Softmax输出应在[0,1]范围内，行%d列%d的值: %f", i, j, val)
+			}
+			
+			rowSum += val
+		}
+		
+		if rowSum < 0.99 || rowSum > 1.01 {
+			t.Errorf("行%d的Softmax值之和应为1，实际值: %f", i, rowSum)
+		}
+	}
+}
+
+func TestFlatten(t *testing.T) {
+	data := []float64{1, 2, 3, 4}
+	shape := []int{2, 2}
+	
+	tensor, err := NewTensor(data, shape)
+	if err != nil {
+		t.Fatal(err)
+	}
+	
+	flattened := tensor.Flatten()
+	
+	// 验证展平后的形状
+	if flattened.Size() != 4 {
+		t.Errorf("展平后的大小应为4，实际值: %d", flattened.Size())
+	}
+	
+	// 验证展平后的值
+	for i := 0; i < 4; i++ {
+		expected := float64(i + 1)
+		actual, err := flattened.Get(i)
+		if err != nil {
+			t.Fatal(err)
+		}
+		
+		if actual != expected {
+			t.Errorf("位置%d的值不正确，期望: %f, 实际: %f", i, expected, actual)
+		}
+	}
+}
+
+func TestMeanSquaredError(t *testing.T) {
+	data1 := []float64{1.0, 2.0, 3.0, 4.0}
+	data2 := []float64{2.0, 3.0, 4.0, 5.0}
+	shape := []int{2, 2}
+	
+	tensor1, err := NewTensor(data1, shape)
+	if err != nil {
+		t.Fatal(err)
+	}
+	
+	tensor2, err := NewTensor(data2, shape)
+	if err != nil {
+		t.Fatal(err)
+	}
+	
+	mse := tensor1.MeanSquaredError(tensor2)
+	
+	// 计算期望的MSE值
+	// ( (1-2)^2 + (2-3)^2 + (3-4)^2 + (4-5)^2 ) / 4 = (1+1+1+1)/4 = 1
+	expected_mse := 1.0
+	actual_mse, err := mse.Data.Get(0)
+	if err != nil {
+		t.Fatal(err)
+	}
+	
+	if actual_mse != expected_mse {
+		t.Errorf("MSE值不正确，期望: %f, 实际: %f", expected_mse, actual_mse)
+	}
+}
+
+func TestMaxPool2D(t *testing.T) {
+	// 创建一个简单的4x4输入，批大小为1，通道数为1
+	data := []float64{1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16}
+	shape := []int{1, 1, 4, 4} // [batch, channel, height, width]
+	
+	tensor, err := NewTensor(data, shape)
+	if err != nil {
+		t.Fatal(err)
+	}
+	
+	// 2x2池化，步长为2
+	pooled, err := tensor.MaxPool2D(2, 2)
+	if err != nil {
+		t.Fatal(err)
+	}
+	
+	// 验证输出形状
+	expected_shape := []int{1, 1, 2, 2}
+	actual_shape := pooled.Shape()
+	if len(actual_shape) != len(expected_shape) {
+		t.Errorf("池化后的形状不正确，期望长度: %d, 实际长度: %d", len(expected_shape), len(actual_shape))
+	}
+	
+	// 检查池化结果是否正确
+	// 左上角: max(1,2,5,6) = 6
+	// 右上角: max(3,4,7,8) = 8
+	// 左下角: max(9,10,13,14) = 14
+	// 右下角: max(11,12,15,16) = 16
+	expected_vals := []float64{6, 8, 14, 16}
+	for i, expected := range expected_vals {
+		actual, err := pooled.Get(0, 0, i/2, i%2)
+		if err != nil {
+			t.Fatal(err)
+		}
+		
+		if actual != expected {
+			t.Errorf("位置[0,0,%d,%d]的池化值不正确，期望: %f, 实际: %f", i/2, i%2, expected, actual)
+		}
+	}
+}
\ No newline at end of file
diff --git a/layers.go b/layers.go
new file mode 100644
index 0000000..b4e032c
--- /dev/null
+++ b/layers.go
@@ -0,0 +1,481 @@
+package gotensor
+
+import (
+	"git.kingecg.top/kingecg/gomatrix"
+	"math"
+)
+
+// Flatten 将多维张量展平为一维
+func (t *Tensor) Flatten() *Tensor {
+	totalSize := t.Size()
+	flatShape := []int{totalSize}
+	
+	// 创建新数据切片
+	flatData := make([]float64, totalSize)
+	
+	// 将原数据复制到扁平化数组
+	shape := t.Data.Shape()
+	if len(shape) == 2 {
+		rows, cols := shape[0], shape[1]
+		for i := 0; i < rows; i++ {
+			for j := 0; j < cols; j++ {
+				val, _ := t.Data.Get(i, j)
+				flatData[i*cols+j] = val
+			}
+		}
+	} else if len(shape) == 4 {
+		// 假设为 [batch, channel, height, width] 格式
+		b, c, h, w := shape[0], shape[1], shape[2], shape[3]
+		for bi := 0; bi < b; bi++ {
+			for ci := 0; ci < c; ci++ {
+				for hi := 0; hi < h; hi++ {
+					for wi := 0; wi < w; wi++ {
+						srcIdx := ((bi*c+ci)*h+hi)*w + wi
+						val, _ := t.Data.Get(bi, ci, hi, wi)
+						flatData[srcIdx] = val
+					}
+				}
+			}
+		}
+	} else {
+		// 对于其他情况，直接复制数据
+		for i := 0; i < totalSize; i++ {
+			val, _ := t.Data.Get(i)
+			flatData[i] = val
+		}
+	}
+
+	resultMatrix, _ := gomatrix.NewMatrix(flatData, flatShape)
+	
+	output := &Tensor{
+		Data: resultMatrix,
+		Op:   "flatten",
+	}
+	
+	output.Prevs[0] = t
+	output.Num_Prevs = 1
+	
+	output.backwardFunc = func() {
+		if t.Grad != nil {
+			// 将梯度重塑回原始形状
+			originalShape := t.Data.Shape()
+			originalSize := t.Size()
+			
+			// 创建一个临时数组来保存重塑后的梯度
+			reshapedGradData := make([]float64, originalSize)
+			
+			// 根据原始形状复制梯度数据
+			if len(originalShape) == 2 {
+				rows, cols := originalShape[0], originalShape[1]
+				for i := 0; i < rows; i++ {
+					for j := 0; j < cols; j++ {
+						srcIdx := i*cols + j
+						gradVal, _ := output.Grad.Get(srcIdx)
+						reshapedGradData[srcIdx] = gradVal
+					}
+				}
+			} else if len(originalShape) == 4 {
+				b, c, h, w := originalShape[0], originalShape[1], originalShape[2], originalShape[3]
+				for bi := 0; bi < b; bi++ {
+					for ci := 0; ci < c; ci++ {
+						for hi := 0; hi < h; hi++ {
+							for wi := 0; wi < w; wi++ {
+								srcIdx := ((bi*c+ci)*h+hi)*w + wi
+								gradVal, _ := output.Grad.Get(srcIdx)
+								reshapedGradData[srcIdx] = gradVal
+							}
+						}
+					}
+				}
+			} else {
+				for i := 0; i < originalSize; i++ {
+					gradVal, _ := output.Grad.Get(i)
+					reshapedGradData[i] = gradVal
+				}
+			}
+			
+			// 创建重塑后的梯度矩阵
+			reshapedGradMatrix, _ := gomatrix.NewMatrix(reshapedGradData, originalShape)
+			
+			// 将重塑后的梯度加到原张量的梯度上
+			newGrad, _ := t.Grad.Add(reshapedGradMatrix)
+			t.Grad = newGrad
+		}
+	}
+	
+	return output
+}
+
+// CrossEntropy 交叉熵损失函数
+func (t *Tensor) CrossEntropy(target *Tensor) *Tensor {
+	// 计算预测值的概率分布
+	predictions := t.Softmax()
+	
+	// 计算交叉熵损失
+	// loss = -sum(target * log(predictions + epsilon))
+	epsilon := 1e-15 // 防止log(0)
+	
+	predData := predictions.Data
+	targetData := target.Data
+	
+	// 计算log(predictions + epsilon)
+	logPredData := make([]float64, predData.Size())
+	for i := 0; i < predData.Size(); i++ {
+		var val float64
+		var err error
+		
+		shape := predData.Shape()
+		if len(shape) == 1 {
+			val, err = predData.Get(i)
+		} else if len(shape) == 2 {
+			_, cols := shape[0], shape[1]
+			val, err = predData.Get(i/cols, i%cols)
+		} else {
+			// 对于其他情况，简单地按索引获取
+			val, err = predData.Get(i)
+		}
+		
+		if err != nil {
+			continue
+		}
+		
+		logPredData[i] = math.Log(val + epsilon)
+	}
+	logPredMatrix, _ := gomatrix.NewMatrix(logPredData, predData.Shape())
+	
+	// 计算target * log(predictions + epsilon)
+	multiplied, _ := targetData.Multiply(logPredMatrix)
+	
+	// 求和得到损失值
+	var totalLoss float64
+	for i := 0; i < multiplied.Size(); i++ {
+		var val float64
+		var err error
+		
+		shape := multiplied.Shape()
+		if len(shape) == 1 {
+			val, err = multiplied.Get(i)
+		} else if len(shape) == 2 {
+			_, cols := shape[0], shape[1]
+			val, err = multiplied.Get(i/cols, i%cols)
+		} else {
+			// 对于其他情况，简单地按索引获取
+			val, err = multiplied.Get(i)
+		}
+		
+		if err != nil {
+			continue
+		}
+		
+		totalLoss += val
+	}
+	totalLoss = -totalLoss
+	
+	// 创建损失张量（标量）
+	lossValue := []float64{totalLoss}
+	lossMatrix, _ := gomatrix.NewMatrix(lossValue, []int{1})
+	
+	output := &Tensor{
+		Data: lossMatrix,
+		Op:   "cross_entropy",
+	}
+	
+	output.Prevs[0] = t
+	output.Num_Prevs = 1
+	
+	output.backwardFunc = func() {
+		if t.Grad != nil {
+			// 计算交叉熵损失相对于预测值的梯度
+			predProbs := predictions.Data
+			targetVals := target.Data
+			
+			gradData := make([]float64, predProbs.Size())
+			for i := 0; i < predProbs.Size(); i++ {
+				var predVal, targetVal float64
+				var err error
+				
+				predShape := predProbs.Shape()
+				if len(predShape) == 1 {
+					predVal, err = predProbs.Get(i)
+					if err != nil {
+						continue
+					}
+					targetVal, err = targetVals.Get(i)
+					if err != nil {
+						continue
+					}
+				} else if len(predShape) == 2 {
+					_, cols := predShape[0], predShape[1]
+					predVal, err = predProbs.Get(i/cols, i%cols)
+					if err != nil {
+						continue
+					}
+					targetVal, err = targetVals.Get(i/cols, i%cols)
+					if err != nil {
+						continue
+					}
+				} else {
+					predVal, err = predProbs.Get(i)
+					if err != nil {
+						continue
+					}
+					targetVal, err = targetVals.Get(i)
+					if err != nil {
+						continue
+					}
+				}
+				
+				// 交叉熵损失的梯度是 pred - target
+				gradData[i] = (predVal - targetVal) / float64(predProbs.Size())
+			}
+			
+			gradMatrix, _ := gomatrix.NewMatrix(gradData, predProbs.Shape())
+			
+			// 将梯度加到原张量的梯度上
+			newGrad, _ := t.Grad.Add(gradMatrix)
+			t.Grad = newGrad
+		}
+	}
+	
+	return output
+}
+
+// Softmax 激活函数
+func (t *Tensor) Softmax() *Tensor {
+	// 假设输入是二维的，第二维是类别维度
+	shape := t.Data.Shape()
+	
+	if len(shape) != 2 {
+		// 如果不是二维，则将其视为一维（单个样本）
+		origSize := t.Size()
+		shape = []int{1, origSize}
+	}
+	
+	rows, cols := shape[0], shape[1]
+	
+	// 计算softmax，按行进行
+	resultData := make([]float64, t.Size())
+	
+	for r := 0; r < rows; r++ {
+		// 找到当前行的最大值，用于数值稳定性
+		maxVal := -math.Inf(1)
+		for c := 0; c < cols; c++ {
+			var val float64
+			var err error
+			
+			if len(t.Data.Shape()) == 2 {
+				val, err = t.Data.Get(r, c)
+			} else {
+				// 如果原数据不是二维的，我们按顺序取值
+				val, err = t.Data.Get(r*cols+c)
+			}
+			
+			if err != nil {
+				continue
+			}
+			
+			if val > maxVal {
+				maxVal = val
+			}
+		}
+		
+		// 计算指数和
+		expSum := 0.0
+		tempVals := make([]float64, cols)
+		for c := 0; c < cols; c++ {
+			var val float64
+			var err error
+			
+			if len(t.Data.Shape()) == 2 {
+				val, err = t.Data.Get(r, c)
+			} else {
+				// 如果原数据不是二维的，我们按顺序取值
+				val, err = t.Data.Get(r*cols+c)
+			}
+			
+			if err != nil {
+				continue
+			}
+			
+			expVal := math.Exp(val - maxVal) // 减去最大值提高数值稳定性
+			tempVals[c] = expVal
+			expSum += expVal
+		}
+		
+		// 计算softmax值
+		for c := 0; c < cols; c++ {
+			resultData[r*cols+c] = tempVals[c] / expSum
+		}
+	}
+	
+	resultMatrix, _ := gomatrix.NewMatrix(resultData, shape)
+	
+	output := &Tensor{
+		Data: resultMatrix,
+		Op:   "softmax",
+	}
+	
+	output.Prevs[0] = t
+	output.Num_Prevs = 1
+	
+	output.backwardFunc = func() {
+		if t.Grad != nil {
+			// 计算softmax的梯度
+			gradData := make([]float64, t.Size())
+			
+			for r := 0; r < rows; r++ {
+				for c1 := 0; c1 < cols; c1++ {
+					gradSum := 0.0
+					for c2 := 0; c2 < cols; c2++ {
+						var s1, s2 float64
+						var err error
+						
+						if len(resultMatrix.Shape()) == 2 {
+							s1, err = resultMatrix.Get(r, c1)
+							if err != nil {
+								continue
+							}
+							s2, err = resultMatrix.Get(r, c2)
+							if err != nil {
+								continue
+							}
+						} else {
+							s1, err = resultMatrix.Get(r*cols+c1)
+							if err != nil {
+								continue
+							}
+							s2, err = resultMatrix.Get(r*cols+c2)
+							if err != nil {
+								continue
+							}
+						}
+						
+						var ds_din float64
+						if c1 == c2 {
+							ds_din = s1 * (1 - s2)  // softmax导数：si*(1-sj) if i=j
+						} else {
+							ds_din = -s1 * s2         // softmax导数：-si*sj if i!=j
+						}
+						
+						var outGrad float64
+						if len(output.Grad.Shape()) == 2 {
+							outGrad, _ = output.Grad.Get(r, c2)
+						} else {
+							outGrad, _ = output.Grad.Get(r*cols+c2)
+						}
+						gradSum += ds_din * outGrad
+					}
+					gradData[r*cols+c1] = gradSum
+				}
+			}
+			
+			gradMatrix, _ := gomatrix.NewMatrix(gradData, shape)
+			
+			// 将梯度加到原张量的梯度上
+			newGrad, _ := t.Grad.Add(gradMatrix)
+			t.Grad = newGrad
+		}
+	}
+	
+	return output
+}
+
+// MeanSquaredError 均方误差损失函数
+func (t *Tensor) MeanSquaredError(target *Tensor) *Tensor {
+	// 计算预测值和真实值之间的差值
+	diff, _ := t.Data.Subtract(target.Data)
+	
+	// 计算平方差
+	squaredDiff, _ := diff.Multiply(diff)
+	
+	// 计算均值
+	var total float64
+	for i := 0; i < squaredDiff.Size(); i++ {
+		var val float64
+		var err error
+		
+		shape := squaredDiff.Shape()
+		if len(shape) == 1 {
+			val, err = squaredDiff.Get(i)
+		} else if len(shape) == 2 {
+			_, cols := shape[0], shape[1]
+			val, err = squaredDiff.Get(i/cols, i%cols)
+		} else {
+			// 对于其他情况，简单地按索引获取
+			val, err = squaredDiff.Get(i)
+		}
+		
+		if err != nil {
+			continue
+		}
+		
+		total += val
+	}
+	
+	mseVal := total / float64(squaredDiff.Size())
+	mseData := []float64{mseVal}
+	mseMatrix, _ := gomatrix.NewMatrix(mseData, []int{1})
+	
+	output := &Tensor{
+		Data: mseMatrix,
+		Op:   "mse",
+	}
+	
+	output.Prevs[0] = t
+	output.Num_Prevs = 1
+	
+	output.backwardFunc = func() {
+		if t.Grad != nil {
+			// MSE损失函数的梯度是 2*(prediction - target)/N
+			gradData := make([]float64, t.Size())
+			n := float64(t.Size())
+			
+			for i := 0; i < t.Size(); i++ {
+				var predVal, targetVal float64
+				var err error
+				
+				shape := t.Data.Shape()
+				if len(shape) == 1 {
+					predVal, err = t.Data.Get(i)
+					if err != nil {
+						continue
+					}
+					targetVal, err = target.Data.Get(i)
+					if err != nil {
+						continue
+					}
+				} else if len(shape) == 2 {
+					_, cols := shape[0], shape[1]
+					predVal, err = t.Data.Get(i/cols, i%cols)
+					if err != nil {
+						continue
+					}
+					targetVal, err = target.Data.Get(i/cols, i%cols)
+					if err != nil {
+						continue
+					}
+				} else {
+					// 对于其他情况
+					predVal, err = t.Data.Get(i)
+					if err != nil {
+						continue
+					}
+					targetVal, err = target.Data.Get(i)
+					if err != nil {
+						continue
+					}
+				}
+				
+				gradData[i] = 2 * (predVal - targetVal) / n
+			}
+			
+			gradMatrix, _ := gomatrix.NewMatrix(gradData, t.Data.Shape())
+			
+			// 将梯度加到原张量的梯度上
+			newGrad, _ := t.Grad.Add(gradMatrix)
+			t.Grad = newGrad
+		}
+	}
+	
+	return output
+}
\ No newline at end of file
diff --git a/tensor.go b/tensor.go
index c80fc54..5938cfc 100644
--- a/tensor.go
+++ b/tensor.go
@@ -1,6 +1,9 @@
 package gotensor
 
-import "git.kingecg.top/kingecg/gomatrix"
+import (
+	"git.kingecg.top/kingecg/gomatrix"
+	"math"
+)
 
 const (
 	Max_Prevs = 10
@@ -295,6 +298,233 @@ func (t *Tensor) String() string {
 	return t.Data.String()
 }
 
+// Sigmoid 激活函数
+func (t *Tensor) Sigmoid() *Tensor {
+	// 计算每个元素的sigmoid值
+	inputShape := t.Data.Shape()
+	resultData := make([]float64, t.Data.Size())
+	
+	// 遍历所有元素计算sigmoid
+	idx := 0
+	if len(inputShape) == 1 {
+		for i := 0; i < inputShape[0]; i++ {
+			val, _ := t.Data.Get(i)
+			resultData[idx] = sigmoid(val)
+			idx++
+		}
+	} else if len(inputShape) == 2 {
+		rows, cols := inputShape[0], inputShape[1]
+		for i := 0; i < rows; i++ {
+			for j := 0; j < cols; j++ {
+				val, _ := t.Data.Get(i, j)
+				resultData[idx] = sigmoid(val)
+				idx++
+			}
+		}
+	} else if len(inputShape) == 4 {
+		batch, ch, h, w := inputShape[0], inputShape[1], inputShape[2], inputShape[3]
+		for b := 0; b < batch; b++ {
+			for c := 0; c < ch; c++ {
+				for h_idx := 0; h_idx < h; h_idx++ {
+					for w_idx := 0; w_idx < w; w_idx++ {
+						val, _ := t.Data.Get(b, c, h_idx, w_idx)
+						resultData[idx] = sigmoid(val)
+						idx++
+					}
+				}
+			}
+		}
+	}
+	
+	result, _ := gomatrix.NewMatrix(resultData, inputShape)
+
+	output := &Tensor{
+		Data: result,
+		Op:   "sigmoid",
+	}
+
+	output.Prevs[0] = t
+	output.Num_Prevs = 1
+
+	output.backwardFunc = func() {
+		if t.Grad != nil {
+			// Sigmoid的导数: sigmoid(x) * (1 - sigmoid(x))
+			gradData := make([]float64, t.Data.Size())
+			gradIdx := 0
+			
+			if len(inputShape) == 1 {
+				for i := 0; i < inputShape[0]; i++ {
+					sigmoidVal, _ := result.Get(i)
+					gradData[gradIdx] = sigmoidVal * (1 - sigmoidVal)
+					gradIdx++
+				}
+			} else if len(inputShape) == 2 {
+				rows, cols := inputShape[0], inputShape[1]
+				for i := 0; i < rows; i++ {
+					for j := 0; j < cols; j++ {
+						sigmoidVal, _ := result.Get(i, j)
+						gradData[gradIdx] = sigmoidVal * (1 - sigmoidVal)
+						gradIdx++
+					}
+				}
+			} else if len(inputShape) == 4 {
+				batch, ch, h, w := inputShape[0], inputShape[1], inputShape[2], inputShape[3]
+				for b := 0; b < batch; b++ {
+					for c := 0; c < ch; c++ {
+						for h_idx := 0; h_idx < h; h_idx++ {
+							for w_idx := 0; w_idx < w; w_idx++ {
+								sigmoidVal, _ := result.Get(b, c, h_idx, w_idx)
+								gradData[gradIdx] = sigmoidVal * (1 - sigmoidVal)
+								gradIdx++
+							}
+						}
+					}
+				}
+			}
+			
+			gradMatrix, _ := gomatrix.NewMatrix(gradData, inputShape)
+			
+			// 与输出梯度相乘
+			gradWithOutput, _ := gradMatrix.Multiply(output.Grad)
+			grad, _ := t.Grad.Add(gradWithOutput)
+			t.Grad = grad
+		}
+	}
+
+	return output
+}
+
+// ReLU 激活函数
+func (t *Tensor) ReLU() *Tensor {
+	// 计算每个元素的ReLU值
+	inputShape := t.Data.Shape()
+	resultData := make([]float64, t.Data.Size())
+	
+	// 遍历所有元素计算ReLU
+	idx := 0
+	if len(inputShape) == 1 {
+		for i := 0; i < inputShape[0]; i++ {
+			val, _ := t.Data.Get(i)
+			if val > 0 {
+				resultData[idx] = val
+			} else {
+				resultData[idx] = 0
+			}
+			idx++
+		}
+	} else if len(inputShape) == 2 {
+		rows, cols := inputShape[0], inputShape[1]
+		for i := 0; i < rows; i++ {
+			for j := 0; j < cols; j++ {
+				val, _ := t.Data.Get(i, j)
+				if val > 0 {
+					resultData[idx] = val
+				} else {
+					resultData[idx] = 0
+				}
+				idx++
+			}
+		}
+	} else if len(inputShape) == 4 {
+		batch, ch, h, w := inputShape[0], inputShape[1], inputShape[2], inputShape[3]
+		for b := 0; b < batch; b++ {
+			for c := 0; c < ch; c++ {
+				for h_idx := 0; h_idx < h; h_idx++ {
+					for w_idx := 0; w_idx < w; w_idx++ {
+						val, _ := t.Data.Get(b, c, h_idx, w_idx)
+						if val > 0 {
+							resultData[idx] = val
+						} else {
+							resultData[idx] = 0
+						}
+						idx++
+					}
+				}
+			}
+		}
+	}
+	
+	result, _ := gomatrix.NewMatrix(resultData, inputShape)
+
+	output := &Tensor{
+		Data: result,
+		Op:   "relu",
+	}
+
+	output.Prevs[0] = t
+	output.Num_Prevs = 1
+
+	output.backwardFunc = func() {
+		if t.Grad != nil {
+			// ReLU的导数: 输入大于0时为1，否则为0
+			gradData := make([]float64, t.Data.Size())
+			gradIdx := 0
+			
+			if len(inputShape) == 1 {
+				for i := 0; i < inputShape[0]; i++ {
+					val, _ := t.Data.Get(i)
+					if val > 0 {
+						gradData[gradIdx] = 1
+					} else {
+						gradData[gradIdx] = 0
+					}
+					gradIdx++
+				}
+			} else if len(inputShape) == 2 {
+				rows, cols := inputShape[0], inputShape[1]
+				for i := 0; i < rows; i++ {
+					for j := 0; j < cols; j++ {
+						val, _ := t.Data.Get(i, j)
+						if val > 0 {
+							gradData[gradIdx] = 1
+						} else {
+							gradData[gradIdx] = 0
+						}
+						gradIdx++
+					}
+				}
+			} else if len(inputShape) == 4 {
+				batch, ch, h, w := inputShape[0], inputShape[1], inputShape[2], inputShape[3]
+				for b := 0; b < batch; b++ {
+					for c := 0; c < ch; c++ {
+						for h_idx := 0; h_idx < h; h_idx++ {
+							for w_idx := 0; w_idx < w; w_idx++ {
+								val, _ := t.Data.Get(b, c, h_idx, w_idx)
+								if val > 0 {
+									gradData[gradIdx] = 1
+								} else {
+									gradData[gradIdx] = 0
+								}
+								gradIdx++
+							}
+						}
+					}
+				}
+			}
+			
+			gradMatrix, _ := gomatrix.NewMatrix(gradData, inputShape)
+			
+			// 与输出梯度相乘
+			gradWithOutput, _ := gradMatrix.Multiply(output.Grad)
+			grad, _ := t.Grad.Add(gradWithOutput)
+			t.Grad = grad
+		}
+	}
+
+	return output
+}
+
+// sigmoid 计算sigmoid函数值
+func sigmoid(x float64) float64 {
+	if x > 0 {
+		z := math.Exp(-x)
+		return 1 / (1 + z)
+	} else {
+		z := math.Exp(x)
+		return z / (1 + z)
+	}
+}
+
 // Backward 执行反向传播
 func (t *Tensor) Backward() {
 	if t.backwardFunc != nil {