Cory Shain | CDR(NN)

In many real world time series, events trigger “ripples” in a dependent variable that unfold slowly and overlap in time (temporal diffusion). Recovering the underlying dynamics of temporally diffuse effects is challenging when events and/or responses occur at irregular intervals. Continuous-time deconvolutional regression (CDR) is a regression technique for time series that directly models temporal diffusion of effects (Shain & Schuler, 2018, 2021) as a funtion of continuous time. CDR uses machine learning to estimate continuous-time impulse response functions (IRFs) that mediate between predictors (event properties) and responses. Given data and a model template specifying the functional form(s) of the IRF kernel(s), CDR finds IRF parameters that optimize some objective function. This approach can be generalized to account for non-stationary, non-linear, non-additive, and context-dependent response functions by implementing the IRF as a deep neural network (CDRNN; Shain, 2021).

Related publications

ACL

CDRNN: Discovering complex dynamics in human language processing

Shain, Cory

In Proceedings of the Joint Conference of the 59th Annual Meeting of the Association for Computational Linguistics and the 11th International Joint Conference on Natural Language Processing 2021

Bib PDF Supp Code Data Slides recording

@inproceedings{shain21,
  author = {Shain, Cory},
  title = {CDRNN: Discovering complex dynamics in human language processing},
  booktitle = {Proceedings of the Joint Conference of the 59th Annual Meeting of the Association for Computational Linguistics and the 11th International Joint Conference on Natural Language Processing},
  year = {2021},
  pdf = {shain_21.pdf},
  slides = {shain_21_slides.pdf},
  code = {https://github.com/coryshain/cdr},
  data = {https://osf.io/hb5w2/},
  supp = {https://osf.io/z89vn/},
  recording = {https://screencast-o-matic.com/watch/cr13ocV13bR},
  abbr = {ACL},
  section = {papers},
  project = {cdrnn},
  bibtex_show = {true}
}

Cognition

Continuous-time deconvolutional regression for psycholinguistic modeling

Shain, Cory, and Schuler, William

Cognition 2021

Bib PDF Code Data

@article{shainschuler21,
  title = {Continuous-time deconvolutional regression for psycholinguistic modeling},
  author = {Shain, Cory and Schuler, William},
  year = {2021},
  journal = {Cognition},
  pages = {104735},
  pdf = {shain_schuler_21.pdf},
  code = {https://github.com/coryshain/cdr},
  data = {https://osf.io/hb5w2/},
  abbr = {Cognition},
  bibtex_show = {true},
  section = {articles},
  project = {cdrnn}
}

Npsy

fMRI reveals language-specific predictive coding during naturalistic sentence comprehension

Shain, Cory, Blank, Idan, van Schijndel, Marten, Schuler, William, and Fedorenko, Evelina

Neuropsychologia 2020

Bib PDF Code Data

@article{shainetal20,
  title = {fMRI reveals language-specific predictive coding during naturalistic sentence comprehension},
  author = {Shain, Cory and Blank, Idan and {van Schijndel}, Marten and Schuler, William and Fedorenko, Evelina},
  year = {2020},
  journal = {Neuropsychologia},
  volume = {138},
  pages = {107307},
  doi = {10.1101/717512},
  pdf = {shain_blank_et_al_20.pdf},
  section = {articles},
  code = {https://github.com/coryshain/cdr},
  data = {https://osf.io/eyp8q/},
  abbr = {Npsy},
  selected = {true},
  project = {cdrnn,mempred},
  bibtex_show = {true}
}

NAACL

A large-scale study of the effects of word frequency and predictability in naturalistic reading

Shain, Cory

In Proceedings of the 2019 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, Volume 1 (Long and Short Papers) 2019

Bib PDF Code Slides recording

@inproceedings{shain19,
  author = {Shain, Cory},
  title = {A large-scale study of the effects of word frequency and predictability in naturalistic reading},
  booktitle = {Proceedings of the 2019 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, Volume 1 (Long and Short Papers)},
  year = {2019},
  pages = {4086--4094},
  doi = {10.18653/v1/N19-1413},
  pdf = {shain_19.pdf},
  slides = {shain_19_slides.pdf},
  code = {https://github.com/coryshain/cdr},
  recording = {https://vimeo.com/359727110},
  abbr = {NAACL},
  section = {papers},
  project = {cdrnn,mempred},
  bibtex_show = {true}
}

EMNLP

Deconvolutional time series regression: A technique for modeling temporally diffuse effects

Shain, Cory, and Schuler, William

In Proceedings of the 2018 Conference on Empirical Methods in Natural Language Processing 2018

Bib PDF Code Poster

@inproceedings{shainschuler18,
  title = {Deconvolutional time series regression: A technique for modeling temporally diffuse effects},
  author = {Shain, Cory and Schuler, William},
  booktitle = {Proceedings of the 2018 Conference on Empirical Methods in Natural Language Processing},
  year = {2018},
  pages = {2679--2689},
  doi = {10.18653/v1/D18-1288},
  pdf = {shain_schuler_18.pdf},
  poster = {shain_schuler_18_poster.pdf},
  code = {https://github.com/coryshain/cdr},
  abbr = {EMNLP},
  section = {papers},
  project = {cdrnn},
  bibtex_show = {true}
}