ganler
commented
3 years ago Mutation rules
Intact-layer mutation
To change the model's topology;
e.g., layer removal, layer copy, layer addition, layer switch (i.e., but the same input/output shape), multi-layer-addition (requires consistent i/o shape), remove/add activation functions...
Inner-layer mutation
- add noise to weights;
- exchange weights;
- neuron activation inverse;
- neuron effect blk;
- neuron switch...
- zero weights...
What to detect
- the number of inconsistent results;
- the number of bugs;
https://www.youtube.com/watch?v=i3tHdZWTifg https://drive.google.com/file/d/16bd0Cs1L02Ejy2LGo8LNVvivhIhJxI0v/view code: https://github.com/Jacob-yen/LEMON