SPAA

Repository containing the code for Spiking Probabilistic Adversarial Attacks.

Setup

• Install cleverhans using pip install git+https://github.com/cleverhans-lab/cleverhans.git#egg=cleverhans
• Install Sinabs pip install sinabs
• Install pytorch pip install torch ujson gdown matplotlib tonic
• Install aermanager pip install aermanager
• Install sinabs-dynapcnn pip install sinabs-dynapcnn==0.2.1.dev53

Tutorial

Start by executing python tutorial_NMNIST.py to see if everything works fine.

Setting up a new experiment

See the Experiments/example_experiment.py. Use quick_access.py to call the experiment when you want to debug for example. Also see the @cachable function in experiment_utils.py for an example of how to implement functions for experiments.

How to run dynapcnn_test.py experiment on the chip

1. Install the latest samna: pip install samna --index-url https://gitlab.com/api/v4/projects/27423070/packages/pypi/simple -U
2. Install sinabs on the master branch. Typically pip install sinabs should do the trick here, otherwise pull the repo manually and install.
3. Install the hardware backend repository for dynapcnndevkit/speck2b sinabs-dynapcnn on the add_reset_method branch. If that branch does not exist, it has probably been merged into master.
4. In the dynapcnn_test.py script, change the variable DYNAPCNN_HARDWARE to your hardware model, e.g. dynapcnndevkit or speck2b