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.
To create:
Bismit simulates the interactions of the following hierarchical structures of the neocortex:
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.