可以实现两个二阶Tensora和b与tf.expand_dims的并矢（外）积，如下所示

product = tf.expand_dims(tf.expand_dims(a, 0), 1) * tf.expand_dims(tf.expand_dims(b, 2), 3)

如果你只需要两个身份，tf.transpose的重新整形为4秩tf.eye应该更简单。

当你想从产品中创造身份时，这是可能的。
样本，创建一个相同的身份。

import tensorflow as tf

class MyDenseLayer(tf.keras.layers.Layer):
    def __init__(self, num_outputs):
        super(MyDenseLayer, self).__init__()
        self.num_outputs = num_outputs
        
    def build(self, input_shape):
        min_size_init = tf.keras.initializers.RandomUniform(minval=10, maxval=10, seed=None)
        self.kernel = self.add_weight(shape=[int(input_shape[-1]),
                        self.num_outputs],
                        initializer = min_size_init,
                        trainable=True)

    def call(self, inputs):
        return tf.matmul(inputs, self.kernel)
        
start = 3
limit = 9
delta = 3
        
# Create DATA
sample_1 = tf.range(start, limit, delta)
sample_1 = tf.cast( sample_1, dtype=tf.float32 )

start = 4
limit = 12
delta = 4

# Create DATA
sample_2 = tf.range(start, limit, delta)
sample_2 = tf.cast( sample_2, dtype=tf.float32 )

# Multiply
sample_1 = tf.constant( sample_1, shape=( 1, 2 ) )
sample_2 = tf.constant( sample_2, shape=( 1, 2 ) )
layer = MyDenseLayer(4)
data = layer(sample_1)
data = layer(sample_2)

print( data )