Connectionists: Learned multidimensional indexes via sparse distributed representations

[Rod Rinkus]

 Dear Connectionists,

I posted a Medium article "Learned Multidimensional Indexes
<https://medium.com/@rod_83597/learned-multidimensional-indexes-171c93fb581a>"
that I believe will be of interest to this community. It explains how when
items are represented as sets (i.e., entities which formally have
extension) as opposed to representing them as points (as is the case for
both localist representations and dense distributed representations, i.e.,
vectors of reals), the items can simultaneously be physically ordered on
multiple uncorrelated underlying dimensions (latent variables).  This
allows immediate best-match retrieval (approximate nearest neighbors) on
any of those uncorrelated, even completely anti-correlated, dimensions.

The essential insight is that the pattern of intersections over the sets
[e.g., sparse distributed representations (SDRs)] constitute "internal
indexes", i.e., internal to the codes themselves, which can be learned
directly from the data.  Such internal indexes remove the need for the
*external
*indexes currently needed for conventional databases.

I look forward to thoughts / comments from the community.

Sincerely
Rod Rinkus


-- 
Gerard (Rod) Rinkus, PhD
President and Chief Scientist
rod at neurithmicsystems dot com
Neurithmic Systems LLC <http://sparsey.com>
Newton, MA 02465
617-997-6272

Visiting Scientist
Volen Center for Complex Systems
Brandeis University, Waltham, MA
grinkus at brandeis dot edu
http://people.brandeis.edu/~grinkus/
<http://people.brandeis.edu/%7Egrinkus/>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mailman.srv.cs.cmu.edu/pipermail/connectionists/attachments/20190213/cd188992/attachment.html>