Cerebra is an engine for developing artificial neural networks and processing them using rules given by natural language. Employing bleeding-edge research in quantum logic (see e.g. this paper ), and optimized using mesh refinement within the spectral Julia set representing chaotic dynamics, it calculates probabilities of a massive field, rather than of single elements, subsequently applying an eigenstate method to interpret abstract, stropped compositions of grammatical elements given a user-defined "spectral" basis of elementary adjectives. This linguistic feature gives it the advantage of being able to index elements by arbitrary properties with speed nearing that of a simple O(k) array lookup.

Currently, the project supports threading, via both stackless python and Nvidia CUDA, and distribution over networks via IPv4. Input is via xml and/or xml syntax trees and numpy arrays. Saving and loading simulations is indirectly supported via pickling and non-blocking file i/o.