tensorflow - 如何在 tensorflow 2.0 中计算 hessian？

Question

https://www.tensorflow.org/api_docs/python/tf/hessians

典型的做法应该是这样

with tf.GradientTape() as tape_:
    with tf.GradienTape() as tape:
        loss = ...
    g = tape.gradient(loss, [vars])
gg = tape_.gradient(loss, [tf.transpose(vars)])

当然，移调不能像磁带那样工作。

tf.hessian 在文档中没有示例。我认为它可能来自 tf 1.0。

更新:使用tf.jacobian

with tf.GradientTape() as tape_:
    with tf.GradientTape() as tape:
        loss = J(a_orig, r)
    dJda = tape.jacobian(loss, [a_orig])[0]
s = tape_.jacobian(dJda, [a_orig])[0]

Answer 1

如更新中所列。

再次使用同一个变量进行微分时，使用tf.jacobian代替tf.gradient。

with tf.GradientTape() as tape_:
    with tf.GradientTape() as tape:
        loss = J(a_orig, r)
    dJda = tape.jacobian(loss, [a_orig])[0]
s = tape_.jacobian(dJda, [a_orig])[0]