DataPoisoning_FL icon indicating copy to clipboard operation
DataPoisoning_FL copied to clipboard

Published 20 hours ago • verified publisher icon git-disl

here, Malicious participants are not visible with mentioned settings. How to get output as in paper?

Open AriesQa opened this issue 3 years ago • 10 comments

here, Malicious participants are not visible with mentioned settings. How to get output as in paper?

defense_results

Originally posted by @AriesQa in https://github.com/git-disl/DataPoisoning_FL/issues/1#issuecomment-858541844

AriesQa avatar Jun 13 '21 13:06 AriesQa

Hi, can you please post your configuration information for your experiments? I suspect this is due to some incorrect configuration.

DarkmatterVale avatar Jun 13 '21 17:06 DarkmatterVale

hi, I have put path, but the result is not same as in paper.

Paths you need to put in.

MODELS_PATH = "/home/XYZ/Desktop/DataPoisoning_FL/3000_models" EXP_INFO_PATH = "/home/XYZ/Desktop/DataPoisoning_FL/logs/3000.log"

kindly tell me how to put POISONED_WORKER_IDS = [] manually?

AriesQa avatar Jun 14 '21 05:06 AriesQa

Hi, can you please post your configuration information for your experiments? I suspect this is due to some incorrect configuration.

Ubuntu 20.04.2.0 LTS Python 3.8 PyTorch 1.8.1 torchvision 0.9.1 loguru 0.3.2 matplotlib 3.4.2 scikit-learn 0.24.2 numpy 1.19.5

AriesQa avatar Jun 17 '21 06:06 AriesQa

In the arguments configuration, you can specify the number of poisoned workers that you would like to simulate in your experiment. You will need to set this attribute to a number larger than zero in order to see results similar to what I demonstrate in the paper.

With regards to identifying the poisoned worker IDs, these are available in the .log file generated for each experiment, near the top. You should see a line that looks something like: Poisoned worker IDs: [...]. The list of poisoned workers is available there, and you can copy and paste that into the defense script file.

DarkmatterVale avatar Jun 19 '21 18:06 DarkmatterVale

In the arguments configuration, you can specify the number of poisoned workers that you would like to simulate in your experiment. You will need to set this attribute to a number larger than zero in order to see results similar to what I demonstrate in the paper.

With regards to identifying the poisoned worker IDs, these are available in the .log file generated for each experiment, near the top. You should see a line that looks something like: Poisoned worker IDs: [...]. The list of poisoned workers is available there, and you can copy and paste that into the defense script file.

Thanks a lot for your response :) I have run with defense.py:29: change it to CLASS_NUM = 1 I have tried it for 10 poisoned workers. Here is the result but scattered. defense_results

AriesQa avatar Jun 20 '21 04:06 AriesQa

Hi, I wonder that have you got the correct result?

mars-ch avatar Apr 19 '22 08:04 mars-ch

Hi, I did not get same results as in paper.

AriesQa avatar Apr 28 '22 12:04 AriesQa

Hi all,

There are a number of reasons why you might be getting this output. In order to diagnose this further, I would need to see the complete configuration of the experiments, along with the poisoned worker ids, and non-poisoned worker IDs.

DarkmatterVale avatar Apr 28 '22 14:04 DarkmatterVale

Hi, should I use my own defense model? This file doesn't have a defense model.

freedom22222 avatar Apr 29 '22 05:04 freedom22222

Hi,

I got a result that looks great.

defense_results

I just modify the parameters that are mentioned in Readme, delete all result files and do some changes mentioned in the closed issue.

mars-ch avatar Apr 29 '22 06:04 mars-ch