Skip to content
/ lamp Public

LAMP: Extracting Text from Gradients with Language Model Priors (NeurIPS '22)

License

Notifications You must be signed in to change notification settings

eth-sri/lamp

Repository files navigation

LAMP: Extracting Text from Gradients with
Language Model Priors portfolio_view

The code accompanying our NeurIPS 2022 paper: LAMP: Extracting Text from Gradients with Language Model Priors.

For a brief overview, check out our blogpost.

Prerequisites

  • Install Anaconda.
  • Create the conda environment:

conda env create -f environment.yml

  • Enable the created environment:

conda activate lamp

  • Download required files:

wget -r -np -R "index.html*" https://files.sri.inf.ethz.ch/lamp/
mv files.sri.inf.ethz.ch/lamp/* ./
rm -rf files.sri.inf.ethz.ch

Main experiments (Table 1)

Parameters

  • DATASET - the dataset to use. Must be one of cola, sst2, rotten_tomatoes.
  • BERT_PATH - the language model to attack. Must be one of bert-base-uncased, huawei-noah/TinyBERT_General_6L_768D, models/bert-base-finetuned-cola, models/bert-base-finetuned-sst2, models/bert-base-finetuned-rottentomatoes for BERTBASE, TinyBERT6, and each of the three fine-tuned BERTBASE-FT models (on each of the datasets).

Commands

  • To run the experiment on LAMP with cosine loss:

./lamp_cos.sh BERT_PATH DATASET 1

  • To run the experiment on LAMP with cosine loss on BERTLARGE:

./lamp_cos_large.sh DATASET 1

  • To run the experiment on LAMP with L1+L2 loss:

./lamp_l1l2.sh BERT_PATH DATASET 1

  • To run the experiment on LAMP with L1+L2 loss on BERTLARGE:

./lamp_l1l2_large.sh DATASET 1

  • To run the experiment on TAG:

./tag.sh BERT_PATH DATASET 1

  • To run the experiment on TAG on BERTLARGE:

./tag_large.sh DATASET 1

  • To run the experiment on DLG:

./dlg.sh BERT_PATH DATASET 1

  • To run the experiment on DLG on BERTLARGE:

./dlg_large.sh DATASET 1

Batch size experiments (Table 2)

Parameters

  • DATASET - the dataset to use. Must be one of cola, sst2, rotten_tomatoes.
  • BATCH_SIZE - the batch size to use e.g 4.

Commands

  • To run the experiment on LAMP with cosine loss:

./lamp_cos.sh bert-base-uncased DATASET BATCH_SIZE

  • To run the experiment on LAMP with L1+L2 loss:

./lamp_l1l2.sh bert-base-uncased DATASET BATCH_SIZE

  • To run the experiment on TAG:

./tag.sh bert-base-uncased DATASET BATCH_SIZE

  • To run the experiment on DLG:

./dlg.sh bert-base-uncased DATASET BATCH_SIZE

Ablation study (Table 4)

Parameters

  • DATASET - the dataset to use. Must be one of cola, sst2, rotten_tomatoes.

Commands

  • To run the ablation experiments in Table 4:

./ablation.sh DATASET

Gaussian noise defense (Table 5)

Parameters

  • SIGMA - the amount of Gaussian noise with which to defend e.g 0.001.

Commands

  • To run the experiment on LAMP with cosine loss:

./lamp_cos.sh bert-base-uncased cola 1 --defense_noise SIGMA

  • To run the experiment on LAMP with L1+L2 loss:

./lamp_l1l2.sh bert-base-uncased cola 1 --defense_noise SIGMA

  • To run the experiment on TAG:

./tag.sh bert-base-uncased cola 1 --defense_noise SIGMA

  • To run the experiment on DLG:

./dlg.sh bert-base-uncased cola 1 --defense_noise SIGMA

Zeroed-out gradient entries defense (Table 8)

Parameters

  • ZEROED - the ratio of zeroed out gradient entries e.g 0.75.

Commands

  • To run the experiment on LAMP with cosine loss:

./lamp_cos.sh bert-base-uncased cola 1 --defense_pct_mask ZEROED

  • To run the experiment on LAMP with L1+L2 loss:

./lamp_l1l2.sh bert-base-uncased cola 1 --defense_pct_mask ZEROED

  • To run the experiment on TAG:

./tag.sh bert-base-uncased cola 1 --defense_pct_mask ZEROED

  • To run the experiment on DLG:

./dlg.sh bert-base-uncased cola 1 --defense_pct_mask ZEROED

Fine-tuning BERT with and without defended gradients

Parameters

  • DATASET - the dataset to use. Must be one of cola, sst2, rotten_tomatoes.
  • SIGMA - the amount of Gaussian noise with which to train e.g 0.001. To train without defense set to 0.0.
  • NUM_EPOCHS - for how many epochs to train e.g 2.

Commands

  • To train your own network:

python3 train.py --dataset DATASET --batch_size 32 --noise SIGMA --num_epochs NUM_EPOCHS --save_every 100

The models are stored under finetune/DATASET/noise_SIGMA/STEPS

Citation

@inproceedings{
    balunovic2022lamp,
    title={{LAMP}: Extracting Text from Gradients with Language Model Priors},
    author={Mislav Balunovic and Dimitar Iliev Dimitrov and Nikola Jovanovi{\'c} and Martin Vechev},
    booktitle={Advances in Neural Information Processing Systems},
    editor={Alice H. Oh and Alekh Agarwal and Danielle Belgrave and Kyunghyun Cho},
    year={2022},
    url={https://openreview.net/forum?id=6iqd9JAVR1z}
}

About

LAMP: Extracting Text from Gradients with Language Model Priors (NeurIPS '22)

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published