I am recruiting a PhD student to work with me, Peter Cholak, Anand Pillay, and Andy Yang @pentagonalize on transformers and logic/model theory (or related topics). If you are interested, please email me with "FLaNN" in the subject line!
David Chiang
808 posts
David Chiang
@davidweichiang
Associate Professor of Computer Science at University of Notre Dame. Artificial intelligence, natural language processing, formal language theory.
- I'm looking for a PhD student to work on NLP (esp. morphology, phonology) for documenting endangered languages.
- Accepted at ICML 2023, a new theoretical result on transformer expressivity and formal logic!
- Justin DeBenedetto's paper on representing unordered sets accepted to ICML! Extends DeepSets and reinterprets Transformer position encodings using automata theory and linear algebra. Earlier version:
- The ACL Anthology has added thousands of curly braces to its BibTeX so you don’t have to! (Corrections welcome.)
- New preprint! Dana Angluin, I, and Andy Yang @pentagonalize show that masked hard-attention transformers are exactly equivalent to the star-free regular languages. arxiv.org/abs/2310.13897
- ACL Anthology now brings all your papers together, (hopefully) even if your name has multiple spellings or if you've changed your name, e.g., aclweb.org/anthology/peop…. If your name isn't right, please submit a correction!
- Accepted at #acl2022, a follow-up to @mhahn29's (excellent) TACL paper on Transformers, parity, and attention. arxiv.org/abs/2202.12172
- Accepted as a spotlight at #ICLR2024: @BrianDuSell on a new variant of self-attention based on pushdown automata. Better-than-ever results on context-free languages, better perplexity on PTB.
- Congratulations to Justin DeBenedetto on passing his PhD thesis defense! Justin's thesis is on graph automata, which can be used in neural models for generating natural language text from semantic graphs, or building semantic graphs from natural language text.
- @DarceyNLP’s and my paper “Factor Graph Grammars” accepted as spotlight at #NeurIPS2020. Hyperedge replacement graph grammars for graphical models.
