How goes it?

[jasond...@yahoo.com]

Just trying to see if you have any update news. Very interested in seeing the latest version run :) Thanks!

[Matthew Rodgers]

We have a Java-based, spiking, fully recurrent (graph!!!), time-dependent, biologically based neural network that can perform deep learning (like a DRQN but _much_ different), and we still have one extra "forward pass" to complete.  It is training faster than we hoped for an REM cycle (about 15 minutes).  Whereas a few months ago it was at 30 minutes (a human takes approximately 3 hours).  We were attempting to optimize the training by trying different methods of neuron growth/death based on multiple factors, research on the amount that normally occurs biologically.  Our model allows dynamic adding and removal of Neurons/Dendrites/Axons without messing up the state of the network, which is unique in the field of AI, currently.  Once we have the last forward pass complete, should just be code cleanup of a bunch of random junk functions to just 4 distinct functions for the "consciousness" workflow.  The GUI is relatively basic at this point (redone in Java), currently only processing text input/output and speech recognition... but plan to add back in vision and the 3d face and such.  We've printed lots of parts for the Inmoov robot, but the business location moved in 2017 and that slowed things down.  We have a nice test framework now that tests each component from the bottom, up, whenever we make a modification; we retest all functions of the dendrite, then neuron, then neural network, then reasoning network, then the AI level.  Rarely launching the full GUI for a while.  We also investigated other emerging technologies as replacements for some components, but for now consider them not sufficient, or incapable of adapting to our model.