Publications and Talks

 

Publications

(Notes: 1. If you don’t have access to github or arxiv, contact me for a copy. 2. Google Colaboratory notebooks look messy, but press the “open with colab” button and they sort themselves out.)

Thinking Like Transformers – ICML 2021 — Gail Weiss, Yoav Goldberg, Eran Yahav. PDF // Github (REPL) // Github (Experiments) // Poster // 5 min talk (ICML 2021): slides, talk // 1 hour talk (Edinburgh 2021):  talk / slides

Synthesizing Context-free Grammars from Recurrent Neural Networks – TACAS 2021 —Daniel Yellin, Gail Weiss. PDF // Github // 20 min talk (TACAS 2021): talk and slides

A Formal Hierarchy of RNN Architectures – ACL 2020 — William Merrill, Gail Weiss, Yoav Goldberg, Roy Schwartz, Noah A. Smith, Eran Yahav.  PDF // Blog Post // 10 min talk (ACL 2020): talk

Learning Deterministic Weighted Automata with Queries and Counterexamples – NeurIPS 2019 — Gail Weiss, Yoav Goldberg, Eran Yahav.  PDF // Github // Poster

On the Practical Computational Power of Finite Precision RNNs for Language Recognition – ACL 2018 — Gail Weiss, Yoav Goldberg, Eran Yahav. PDF // 10 min talk (ACL 2018): talk // Github // Google Colaboratory

  • Notes: This paper is mostly an observation that LSTMs can implement a counting mechanism and GRUs can’t, so the code here is just a demonstration of this.

Extracting Automata from Recurrent Neural Networks Using Queries and Counterexamples – ICML 2018 — Gail Weiss, Yoav Goldberg, Eran Yahav. PDF // 10 min talk (ICML 2018): talk // Github // Google Colaboratory // Poster (with thanks to Xiaokun Luan for corrections!) // Extended 2022 Version, open link (Springer)

 

 

Talks (Independent of Publications)

Formal Abstractions of Neural Sequence Models – ICGI 2021 (Tutorial, 1.5hrs) talkslides