Deep Learning Frameworks
Software libraries powering deep learning research and development
PFN develops and provides open-source deep learning libraries that support research and development in advanced, practical deep learning applications. PFN has driven the advancement of deep learning technology since June 2015 when it open-sourced Chainer™, the deep learning framework that became first in the world to adopt the define-by-run approach. Since migrating the company’s deep learning research framework from Chainer to PyTorch in December 2019, PFN has been closely working with Meta as part of the PyTorch developer community and contributing software libraries to the ecosystem.
Contribution to PyTorch Community
PFN contributes Lazy Modules, fast DataLoader and other useful features of pytorch-pfn-extras, a PFN-developed software library, to PyTorch. As part of the PyTorch community, PFN also contributes articles to the official PyTorch blog.
Deep Learning Libraries
Library of utilities that streamline PyTorch-based research
Deep reinforcement learning library for PyTorch
Chainer & Chainer Extensions
Chainer is a Python-based deep learning framework developed and provided by PFN. First open-sourced in June 2015, Chainer was the first to adopt the define-by-run approach that allows developers to build complex neural networks in intuitive and flexible ways. The approach has gained wide support from research and development communities and is now standard in PyTorch and other deep learning frameworks. Chainer moved into a maintenance phase in December 2019 with the last v7 update. Currently, PFN is closely collaborating with the PyTorch community to migrate Chainer’s key elements and features to PyTorch.
Image recognition algorithms and dataset wrappers
Deep reinforcement learning library
ChainerRL ChainerRL – Blog
Visualization and experiment management tool
ChainerUI ChainerUI – Blog
Graph convolutions for biology/chemistry tasks
Chainer Chemistry Chainer Chemistry – Blog