Chainer
libraryDeep learning framework that pioneered the "define-by-run" dynamic computation graph paradigm in 2015, before PyTorch adopted the same approach. Included CuPy (NumPy-like GPU arrays), ChainerMN (distributed training), and ChainerCV. Development ended Dec 2019 when PFN migrated to PyTorch. KDD 2019.
Paper
arXiv: 1908.00213
Venue: KDD 2019