You could change the file path of the dataset and run the code on google colab.
What I have done:
- Filled the “To-do” session in Homework_2_wz2295.ipynb. Produced three h5 files with different required X{2%, 4%, 10%}, and I have uploaded those files.
Apply eval.py on them to get the acc and asr:
X= 2%
X= 4%
X= 10%
- Draw a table with the accuracy on clean test data and the attack success rate (on backdoored test data) as a function of the fraction of channels pruned (X). (In the report)