fpicetti / occamypy Goto Github PK

CNNs can be seen as nonlinear operators.

Two ideas here:

1. use a CNNs as operator where only the forward is implemented; CNN can be used as regularizers (link)
2. implement a VectorTorch method to build a torch.nn.Parameter object, to be passed to torch.optim.

Dear Dr.Picetti @fpicetti

Thanks for sharing tutorials about automatic differentiation.

These tutorials are very interesting.

I have some question.

1.For my understating, I only need define the forwarding operator and cost function with automatic differentiation. Is it right?

2. How can we use automatic differentiation with nonlinear conjugate gradient method?

3.Would you like add some more tutorials about VSP(1D), tomography (2D) inversion with automatic differentiation? I think the implements are very helpful.

Best regards,
Xi

In See here

Sometimes, useful operators are developed in MATLAB. It would be useful to have a Operator interface that execute matlab scripts.

@fpicetti @biondiettore

Dear sir,

Thanks for this great package.

Your tutorial helps me alot.

However, I want invert some value based eikonal equation and it's adjoint state equation.

My goal is to fit Bessel_j0(a*t), where a is a constant, t is travel time.

How to invert it using adjoint state? This problem has plagued me for several months.

Looking forward to your reply.

Best regards,
Xi

This computation can be useful for other applications.

The Alternating Direction Method of Multipliers is a powerful proximal algorithm that can be useful for Generalized LASSO problems.

Create FromPylops and ToPylops linear operators.
Probably the main issue is Dask.

I install your package, but I can not use it. As I import the package in python, it gives this error:
Can you help me to resolve this issue?

 File "C:\Users\Arash\anaconda3\lib\site-packages\occamypy\utils\sep.py", line 8, in <module>
    HOME = os.environ["HOME"]

  File "C:\Users\Arash\anaconda3\lib\os.py", line 675, in __getitem__
    raise KeyError(key) from None

KeyError: 'HOME'

Hi, thanks for sharing this great package.

I have defined my forward problem. Is it nessary to give the function of Jacbian matrix for LeastSquares inversion? And how to define the function of Jacobian matrix?

Hello Dr. @biondiettore

Thanks for sharing this package. This package is very interesting.

And the 2D travel-time tomography tutorial is very attractive.

Is the 2D travel-time tomography suit for cross-hole type tomography?

Best regards,
Xi