Bismit

Biologically Inspired Sensory Motor Inference Toolkit

Bismit is one of the first members of the next paradigm of cortical learning libraries. Going beyond simple Bayesian neural networks and incorporating ideas from the theory of hierarchical temporal memory, such as sparse distributed representations and temporal context, Bismit is a brain building framework for designing truly intelligent machines. It is not a typical machine learning or deep learning platform and does not function like a traditional neural network.

Goals

To create:

• A toolkit that is simple enough to be useful now and be flexible enough to absorb future neuroscientific discoveries. We know enough now about the basic architecture of the brain that we can create a scaffold with enough wiggle room to be refined as more is learned.

Structure

Bismit simulates the interactions of the following hierarchical structures of the neocortex:

Cortex - Region - Area - Layer - Slice - Cell - Tuft - Dendrite - Synapse

Bismit is written in Rust and OpenCL C and is in an unstable pre-alpha stage. Full basic sensory functionality is complete. Motor control and use case development are underway. See the vibi project (currently in early development) for an OpenGL based visualization frontend.