We gratefully acknowledge support from
the Simons Foundation and member institutions.
Full-text links:

Download:

Current browse context:

cs.CL

Change to browse by:

References & Citations

DBLP - CS Bibliography

Bookmark

(what is this?)
CiteULike logo BibSonomy logo Mendeley logo del.icio.us logo Digg logo Reddit logo ScienceWISE logo

Computer Science > Computation and Language

Title: FireBERT: Hardening BERT-based classifiers against adversarial attack

Abstract: We present FireBERT, a set of three proof-of-concept NLP classifiers hardened against TextFooler-style word-perturbation by producing diverse alternatives to original samples. In one approach, we co-tune BERT against the training data and synthetic adversarial samples. In a second approach, we generate the synthetic samples at evaluation time through substitution of words and perturbation of embedding vectors. The diversified evaluation results are then combined by voting. A third approach replaces evaluation-time word substitution with perturbation of embedding vectors. We evaluate FireBERT for MNLI and IMDB Movie Review datasets, in the original and on adversarial examples generated by TextFooler. We also test whether TextFooler is less successful in creating new adversarial samples when manipulating FireBERT, compared to working on unhardened classifiers. We show that it is possible to improve the accuracy of BERT-based models in the face of adversarial attacks without significantly reducing the accuracy for regular benchmark samples. We present co-tuning with a synthetic data generator as a highly effective method to protect against 95% of pre-manufactured adversarial samples while maintaining 98% of original benchmark performance. We also demonstrate evaluation-time perturbation as a promising direction for further research, restoring accuracy up to 75% of benchmark performance for pre-made adversarials, and up to 65% (from a baseline of 75% orig. / 12% attack) under active attack by TextFooler.
Comments: 8 pages, 10 figures, code available at: this https URL
Subjects: Computation and Language (cs.CL); Machine Learning (cs.LG)
Cite as: arXiv:2008.04203 [cs.CL]
  (or arXiv:2008.04203v1 [cs.CL] for this version)

Submission history

From: Gunnar Mein [view email]
[v1] Mon, 10 Aug 2020 15:43:28 GMT (5726kb,D)

Link back to: arXiv, form interface, contact.