MIPVerify.jl issues

Support for ONNX models

1

Dear developers, thank you for the development of such an amazing tool. I would like to know if it is possible to use MIPVerify with a model represented in Open...

cnrdn

feature request

Code coverage for src/net_components/layers/zero.jl

The file `src/net_components/layers/zero.jl` currently has [zero test coverage](https://codecov.io/gh/vtjeng/MIPVerify.jl/src/2f2a0918abe28fb5f8b0b14396363c516a9c80c6/src/net_components/layers/zero.jl). Following the examples in `test/net_components/layers` (e.g. [`relu.jl`](https://github.com/vtjeng/MIPVerify.jl/blob/master/test/net_components/layers/relu.jl)), add test to cover this functionality.

vtjeng

help wanted

good first issue

Performance on CNN

7

How does it perform on a CNN? I have a `.mat` file containing the weights for a standard MNIST CNN classifier. ``` cnn = Sequential([ MIPVerify.get_conv_params(w_dict, "00-Conv2D", (3,3,1,32)), ReLU(), MIPVerify.get_conv_params(w_dict,...

rnbguy

How does maxpooling layer process Gurobi variable?

1

maxpooling will choose the max element from the kernel, but you don't know the specific value of the node, it just gurobi variable or expression. I want to know how...

UserAlreadyTaken

Add an AdversarialExampleObjective: worst_with_constraint

2

I find that in some cases using the `worst` as adversarial example objective results in much faster solving. However the current implementation does not add the constraint to enforce a...

jia-kai

Problem with very sparse network

2

Hello, I am using MIPVerify to test the robustness of very sparse networks (i.e., in which many of the weights are 0 or near to 0, for example, 10^-8). When...

darioguidotti

System Error using "find_adversarial_example"

2

Hello, I was trying to use "find_adversarial_example" on a fully connected network and I have came across this error: """""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""" [notice | MIPVerify]: Attempting to find adversarial example. Neural net...

darioguidotti