The Nengo team is pleased to announce the release of Nengo 2.0, a Python library for building and simulating large-scale neural models. Nengo can create sophisticated neural simulations with sensible defaults in few lines of code, yet is extensible and flexible enough to use spiking and non-spiking neuron types in the same model, get input directly from hardware, drive robots, and simulate models on diverse computing resources.

This is the first release of Nengo that is implemented entirely in Python, and integrates well with tools like Matplotlib and IPython. Currently, it is designed to be used programmatically, but a browser-based graphical interface is under active development.

Features

Nengo has support for models using the following neuron types, which can be combined in the same model.

  • Rectified linear
  • Sigmoid
  • Leaky integrate-and-fire (spiking and non-spiking)
  • Adaptive leaky integrate-and-fire (spiking and non-spiking)
  • Izhikevich
  • Direct mode (in which mathematical functions are computed directly, rather than approximated from neural activity)

Nengo does not require online learning in the form of synaptic weight changes; however, learning rules do exist for situations when the objective function is not known ahead of time. Nengo has support for the following learning rules:

  • PES rule (minimizes an error signal)
  • BCM rule
  • Oja rule

See the documentation for more details and example usage.

Changes in Nengo 2.0

Nengo 2.0 is a completely new code base that implements all commonly used features of Nengo 1.4.

Links