BERTops

Source code for the paper BERTops: Studying BERT Representations under a Topological Lens accepted at IJCNN 2022 as an Oral Presentation. Arxiv Link

Usage

Requirements

Provided in requirements.txt file

Sample usage for SST-2 dataset

To generate the Persistence Diagrams (PDs), run the following sample code

python generate_label_pds.py --dataset_name sst-2 --base_model bert --model_card echarlaix/bert-base-uncased-sst2-acc91.1-d37-hybrid

Here, the dataset_name, base_model and model_card (from Huggingface community models) can be adjusted for the usage.

To compute the PSF value for generated PDs, run the following sample code

python compute_PSF.py --dataset_name sst-2 --base_model bert --model_card echarlaix/bert-base-uncased-sst2-acc91.1-d37-hybrid

To evaluate (compute test accuracy) the model, run the following sample code

python evaluate_test.py --dataset_name sst-2 --base_model bert --model_card echarlaix/bert-base-uncased-sst2-acc91.1-d37-hybrid

To perform adversaril attack over the model (example run with Textbugger Black-box), run the following sample code

python perform_attack.py --dataset_name sst-2 --base_model bert --model_card echarlaix/bert-base-uncased-sst2-acc91.1-d37-hybrid

Citation

If you find this repository helpful, please consider citing our work:

@article{https://doi.org/10.48550/arxiv.2205.00953,
  doi = {10.48550/ARXIV.2205.00953},
  url = {https://arxiv.org/abs/2205.00953},
  author = {Chauhan, Jatin and Kaul, Manohar},
  keywords = {Machine Learning (cs.LG), FOS: Computer and information sciences, FOS: Computer and information sciences},
  title = {BERTops: Studying BERT Representations under a Topological Lens},
  publisher = {arXiv},
  year = {2022},
  copyright = {Creative Commons Attribution 4.0 International}
}