According to the Chainer documentation2:
"Chainer is a flexible framework for neural networks. One major goal is flexibility, so it must enable us to write complex architectures simply and intuitively."
Chainer adopts a define-by-run scheme, that is, the network is defined dynamically via the actual forward computation. More precisely, Chainer stores the history of computation instead of programming logic. For example, Chainer does not need any magic to introduce conditionals and loops into the network definitions. The define-by-run scheme is the core concept of Chainer. This strategy also makes it easy to write multi-GPU parallelization, since logic comes closer to network manipulation.
Kelp.Net can load a Chainer model directly from disk.