Dataflow Matrix Machines as a Generalization of Recurrent Neural Networks
This proposes a new framework for researchers in machine learning and programming, but it is incremental as it builds on existing recurrent neural network concepts.
The paper introduces dataflow matrix machines as a generalization of recurrent neural networks, enabling work with multiple types of linear streams and neurons, and expects them to be useful in machine learning, probabilistic programming, and system synthesis.
Dataflow matrix machines are a powerful generalization of recurrent neural networks. They work with multiple types of arbitrary linear streams, multiple types of powerful neurons, and allow to incorporate higher-order constructions. We expect them to be useful in machine learning and probabilistic programming, and in the synthesis of dynamic systems and of deterministic and probabilistic programs.