Model Composition

Beyond individual models, nnodely supports composition of multiple neural models and fine-grained wiring of signals. Composition can be performed at two distinct levels:

• Stream-level composition : local operations on signal streams (time windows, delays, closed-loop connections). Implemented by the Stream API (methods such as connect(), closedLoop(), delay(), s(), sw(), tw(), z()).

• Model-level composition : composing and connecting independently defined models. Provided by the Composer interface (methods such as addModel(), removeModel(), addConnect(), addClosedLoop(), removeConnection(), and neuralizeModel()).

• Dynamically : connection are created dynamically at execution time, during training, analysis, or inference.

• Stream-level composition
• Model-level composition
• Composition at Training and Inference Time