Functional Neural Architectures

Functional Neural Architectures (FNA) is a python library for neural network functional benchmarking, analysis and comparison. It provides high-level wrapper for PyNEST and TensorFlow (which is used as the core simulation engines). As such, the types of architectures and their properties are determined by the models available in NEST / TF. The use of these simulators allows efficient and highly scalable simulations of very large and complex circuits, constrained only by the computational resources available to the user.

The modular design allows the user to specify numerical experiments with varying degrees of complexity depending on concrete research objectives. The generality of some of these experiments allows the same types of measurements to be performed on a variety of different circuits, which can be useful for benchmarking and comparison purposes. Additionally, the code was designed to allow an effortless migration across computing systems, i.e. the same simulations can be executed in a local machine, in a computer cluster or a supercomputer, with straightforward resource allocation.

The code is licensed under GPLv2.