Testing-based Black-box Extraction of Simple Models from RNNs and Transformers

Edi Muskardin; Martin Tappler; Bernhard K. Aichernig

Testing-based Black-box Extraction of Simple Models from RNNs and Transformers

Edi Muskardin, Martin Tappler, Bernhard K. Aichernig

Trustworthy & Adaptive Computing (TAC)

Research output: Conference proceeding/Chapter in Book/Report/ › Conference Paper › peer-review

Abstract

In this technical report, we outline the testing-based black-box method used to extract simple and interpretable models from RNNs and transformers. Our work was done in the scope of the TAYSIR competition, in which it won the first place.

Original language	English
Title of host publication	International Conference on Grammar Inference
Volume	217
Publication status	Published - Jul 2023

Keywords

Automata Learning
Model extraction
RNN
Transformers

Access to Document

https://proceedings.mlr.press/v217/muskardin23a/muskardin23a.pdf

Cite this

@inproceedings{6df480096d4d4ed1b8f58c9588129661,

title = "Testing-based Black-box Extraction of Simple Models from RNNs and Transformers",

abstract = "In this technical report, we outline the testing-based black-box method used to extract simple and interpretable models from RNNs and transformers. Our work was done in the scope of the TAYSIR competition, in which it won the first place.",

keywords = "Automata Learning, Model extraction, RNN, Transformers",

author = "Edi Muskardin and Martin Tappler and Aichernig, {Bernhard K.}",

year = "2023",

month = jul,

language = "English",

volume = "217",

booktitle = "International Conference on Grammar Inference",

}

TY - GEN

T1 - Testing-based Black-box Extraction of Simple Models from RNNs and Transformers

AU - Muskardin, Edi

AU - Tappler, Martin

AU - Aichernig, Bernhard K.

PY - 2023/7

Y1 - 2023/7

N2 - In this technical report, we outline the testing-based black-box method used to extract simple and interpretable models from RNNs and transformers. Our work was done in the scope of the TAYSIR competition, in which it won the first place.

AB - In this technical report, we outline the testing-based black-box method used to extract simple and interpretable models from RNNs and transformers. Our work was done in the scope of the TAYSIR competition, in which it won the first place.

KW - Automata Learning

KW - Model extraction

KW - RNN

KW - Transformers

M3 - Conference Paper

VL - 217

BT - International Conference on Grammar Inference

ER -

Testing-based Black-box Extraction of Simple Models from RNNs and Transformers

Abstract

Keywords

Access to Document

Fingerprint

Cite this