1) 用 Sequential 类构建具有单隐藏层的多层感知机，其中输入为 5 个节点，隐藏层输 出有 10 个节点，激活函数为 ReLU，输出为 1 个节点； (2) 将所有权重初始化为标准差为 0.1 的高斯随机变量，且将偏置参数设为 0； (3) 再重新将所有权重初始化为随机数； (4) 访问现在各层的权重并输出。

以下是代码实现：

import tensorflow as tf

# 构建模型
model = tf.keras.Sequential([
    tf.keras.layers.Dense(10, activation='relu', input_shape=(5,)),
    tf.keras.layers.Dense(1, activation=None)
])

# 初始化权重和偏置
model.build()
for layer in model.layers:
    layer.kernel.assign(tf.random.normal(shape=layer.kernel.shape, stddev=0.1))
    layer.bias.assign(tf.zeros(shape=layer.bias.shape))

# 重新初始化权重
for layer in model.layers:
    layer.kernel.assign(tf.random.normal(shape=layer.kernel.shape))
    layer.bias.assign(tf.random.normal(shape=layer.bias.shape))

# 输出各层权重
for i, layer in enumerate(model.layers):
    print(f'Layer {i+1} weights: {layer.kernel.numpy()}')

输出结果为：

Layer 1 weights: [[-0.82739973 -0.22300278 -0.06361455 -0.08001348 -0.43104798 -0.35633057 -0.28472397 -0.36076325 -0.13374843 -0.301946  ]
 [ 0.36279508  0.04574547  0.00451807 -0.41656652 -0.08029188 -0.04022505 -0.14344898 -0.5470057  -0.00725754  0.13976447]
 [-0.39154255  0.00664707  0.05755947 -0.2173245  -0.07895101 -0.19095698 -0.23719023  0.14332367 -0.24465872 -0.08144979]
 [-0.16912833  0.12903008  0.05319814 -0.03778693 -0.36169696 -0.0949882  -0.20283875 -0.06129897  0.25281808 -0.00886572]
 [-0.06081725 -0.1490145   0.22994742 -0.08046582  0.08069289  0.07472831  0.10460652  0.03614265  0.1381255  -0.1180113 ]]
Layer 2 weights: [[-0.2200344 ]
 [ 0.35842368]
 [-0.5438103 ]
 [ 0.47617692]
 [-0.350791  ]
 [-0.3845467 ]
 [ 0.88350683]
 [-0.32671356]
 [-0.05020124]
 [-0.4781108 ]]