I need to work with individual Wiener increments from an SDE solution in my loss function. Whenever I try to get these values from sol.W.W command, the ERROR: type Nothing has no field W pops up when computing the gradients. Can it be a bug in DiffEqSensitivity package? MWE: using DifferentialEqua…

I think the problem traces back to sol.W being nothing in the backward pass. Using ForwardDiff instead of Zygote to compute the gradients @time gs = ForwardDiff.gradient(p) do p some_loss(p) end removes the error. If reverse-mode AD is needed, you could use pre-defined noise values and NoiseGrid …

Using the noise parameters sol.W.W from an SDE solution throws an error when taking gradients

Specific Domains Machine Learning

MikeKlocCZ October 7, 2020, 2:13pm 4

Thank you both for comments and suggestions!

Topic		Replies	Views
Sensitivity of an SDE with non-diagonal noise Modelling & Simulations sde , forwarddiff , differentialequation	2	81	April 4, 2025
Sensitivity analysis on Non-diagonal noise SDEs Modelling & Simulations diffeq , sde , sciml	1	370	August 31, 2021
Gradients from an SDE solver with callbacks Machine Learning question , sde	2	528	September 11, 2020
ForwardDiffSensitivity() and TrackerAdjoint() give different answers Machine Learning diffeq , sde	3	819	July 18, 2020
DiffEqGPU.jl with CUDA: Error computing gradients through SDE solver Machine Learning	12	349	January 10, 2024

Using the noise parameters sol.W.W from an SDE solution throws an error when taking gradients

Related topics