Thanks for sharing this Matt and Mirko. I find the witness complexes sharing many similarities with the geospatial encoder. The idea is first to assign a random feature to each location in a metric space, and then define the SDR for each location by the collection of its neighbors. Chetan's slides have some good illustration of how geospatial encoder works [1].
[1] http://chetansurpur.com/slides/2014/8/5/geospatial-encoder.html#1 -- Yuwei Cui, PhD Research Engineer, Numenta Inc. On Tue, Apr 19, 2016 at 8:28 AM, Matthew Taylor <[email protected]> wrote: > Hello NuPIC, > > I just wanted to point out a new paper sent to me by Mirko Klukas: > http://blog.mirkoklukas.com/sparse-distributed-representations-and-witness-complexes/ > > I've added this to our community content wiki: > https://github.com/numenta/nupic/wiki/Community-Content > > Thanks Mirko! > > --------- > Matt Taylor > OS Community Flag-Bearer > Numenta >
