loadingbyte / are-gnn-defenses-robust Goto Github PK

Thank you for sharing your code. I think this work is very nice for the research communities that are interested in the adversarial robustness of graph.

However, I cannot easily reproduce the code due to the implementation based on the MongoDB. Do you have any plan for adding the setting for Mongo DB. Otherwise, can you add a link for setting to use pymongo in your readme.md of the repository?

I cannot resolve the following errors.

pymongo.errors.ServerSelectionTimeoutError: 143.xxx.xx.xx:xxxxx: [Errno 111] Connection refused, Timeout: 30s, Topology Description: <TopologyDescription id: 64a3adb83c180daa30bebe56, topology_type: Unknown, servers: [<ServerDescription ('143.xxx.xx.xx', xxxxx) server_type: Unknown, rtt: None, error=AutoReconnect('143.xxx.xx.xx:xxxxx: [Errno 111] Connection refused')>]>

Thanks for sharing the source code! Could you please talk about something about why Parameter, Modules, and Optimizers instead of native pytorch stuffs are used here?

In the paper the global evasion adaptive attack (PGD) in Soft-Median will initialize PGD with strong perturbation.

So how to initialize?