Extracting Automata from Recurrent Neural Networks Using Queries and Counterexamples

November 27, 2017 · Entered Twilight · 🏛 International Conference on Machine Learning

"Last commit was 6.0 years ago (≥5 year threshold)"

Evidence collected by the PWNC Scanner

Repo contents: DFA.py, Extraction.py, GRU.py, Helper_Functions.py, LSTM.py, LinearTransform.py, Lstar.py, ObservationTable.py, Quantisations.py, README.md, RNNClassifier.py, Specific_Language_Generation.py, Teacher.py, Tomita_Grammars.py, Training_Functions.py, WhiteboxRNNCounterexampleGenerator.py, dfa_from_rnn.ipynb, dfa_from_rnn_no_documentation.ipynb, dfa_from_rnn_notebook_for_several_rnns.ipynb

Authors Gail Weiss, Yoav Goldberg, Eran Yahav arXiv ID 1711.09576 Category cs.LG: Machine Learning Cross-listed cs.FL Citations 205 Venue International Conference on Machine Learning Repository https://github.com/tech-srl/lstar_extraction ⭐ 77 Last Checked 1 month ago

Abstract

We present a novel algorithm that uses exact learning and abstraction to extract a deterministic finite automaton describing the state dynamics of a given trained RNN. We do this using Angluin's L* algorithm as a learner and the trained RNN as an oracle. Our technique efficiently extracts accurate automata from trained RNNs, even when the state vectors are large and require fine differentiation.