Joel's work log January 2009
2009-01-27 to 2009-01-230
- Experimenting with glocal key nodes
- Implemented new palimpsest training (I think I finally understand what Storkey was actually testing now).
- Graphs of previous glocal work Rui Liu and examining code. Unfortunately the training scheme is not really comparable and I've been unable to contact him about getting recent code or what the table of results specifically represent.
- Found that the modular CogServer prevented the ImportanceDiffusionAgent from using the Hyperbolic spread function.
- Fixed Hyperbolic spread decider to use the correctly normalised STI values.
- Playing around with values. Confused about StorkeyUpdate system - -ve values reinforce one another.
e.g. if you have the pattern [ 1 1 0 0 ] the two zeroes reinforce one another, although that's because they are actually stored as -1. Going to try it with 0's actually stored as 0.
- Hopfield network bug fixing/experimenting.
- Glocal hopfield net work
- Spent some time reading about OpenMPI, NUMA (Non uniform memory access), and thinking about how to adapt the AtomSpace API to fit with distributed and local multi-threaded instances.
- OpenMP, the shared API for parallel programming, on the other hand, would be perfect for calculating the lookup tables, doing the indefinite formula calculations, or evaluating quantifiers.
- tracked down cause of segfault by stop-agents for Jared (threading issues).
- fix Link::addOutgoingAtom()
2009-01-11 - 2009-01-13
- convert Glocal memory paper to Latex
- start implementing key nodes in Hopfield network emulator.
- guiding Nil with Indefinite fixes.
- merge molchuka's MSVC9 patch.
- Reading glocal ECAN paper
- investigating Elsevier submission format (there is none?)
- consolidating various notes, reading wiki.