Say I have a model m = Dense(1,1), a function f(x) = x, and a loss(a,b,c) = m(a) - b - f(c). I don’t want zygote to differentiate on f(c) when calling gradient(() -> loss(a,b,c), m) . How do I tell it that ? It is a problem because the function f is generating random numbers and I get weird errors …

Zygote: treating a model output as a constant

simeonschaub February 4, 2020, 5:18pm 3

Are you looking for Zygote.dropgrad? There was some discussion about a similar problem in this thread.

Topic		Replies	Views
Challenges with Zygote Differentiation in Multivariate Mixture Models as input noise of an implicit machine learning method Machine Learning machine-learning , zygote	10	166	April 5, 2024
In Zygote, why (model)->... in gradient Machine Learning	4	775	February 10, 2020
How to make Zygote avoid differentiating with respect to some fields in struct Machine Learning question , machine-learning , zygote , struct , autodiff	6	1545	August 2, 2021
Zygote gradients for functions with Array output General Usage	2	996	May 16, 2020
Zygote gradient General Usage	2	1044	July 24, 2019