Maybe. Note, though, there is something really really important that happens when "word tuples" get replaced by disjuncts. Its hard to talk about because its both "obvious" and totally obscure...
The point is that if you have some complex pattern of things connected to other things, you have this problem of trying to figure out how to count it, which lead to our arguments about "surprisingness" earlier. The whole point of the disjunct is that it replaces the pattern by a kind-of snapshot of it, a building-block. And this simplification also makes the complexities f tuples and patterns "go way", or rather, converts them into something manageable. Again: it decomposes a pattern into the building-blocks of a pattern. I think this means that you can use linear tools on these building blocks, and when you re-assemble the whole pattern, the non-linearity re-emerges. Perhaps one way to think of this is .. well, if you recall what an atlas is, in topology: its a set of flat maps, that you can glue together to get a non-flat manifold. e.g. literally, the earth is round, but maps are flat, you glue them together to get a round earth. So same here: the disjuncts are the flattened parts of a pattern. you can glue them together to get the whole complex pattern, but by working with the flattened pieces, everything becomes much much simpler. So your word-tuples are undoubtedly non-linear: and that is the point: don't work with word-tuples. They are a difficult, bad representation. --linas On Tue, Jun 6, 2017 at 7:27 AM, Ben Goertzel <[email protected]> wrote: > Interesting! > > Clearly this is doing the right sorts of things, so it should do > something in the vicinity of what's needed... > > still tho -- My intuition remains that a more fully nonlinear NN > approach might do better than a "linear algebra plus thresholding" > approach like this... Put differently, I think we need some more > powerful learning method like evolutionary-learning or backprop in > there, to capture the nonlinear dependencies btw word tuples... > > But this seems worth trying and who knows, maybe it will be awesome... > it will be good to compare different approaches... > > ben > > On Tue, Jun 6, 2017 at 2:13 PM, Linas Vepstas <[email protected]> > wrote: > > Ben, > > > > The attached PDF describes the algorithm I plan to implement for > performing > > the actual clustering. As of right now, I really like it: its simple, its > > straightforward, I believe it will work well. It might be a real CPU > > burner, though, and so blue skies might bring tears. > > > > I like to think of it as a kind-of "pattern miner", as it can be made > > completely generic; it works for any correlation matrix. I suspect that > it > > is totally different from what Shujing does, which is still on my list of > > things to study in greater detail. > > > > --linas > > > > -- > > You received this message because you are subscribed to the Google Groups > > "link-grammar" group. > > To unsubscribe from this group and stop receiving emails from it, send an > > email to [email protected]. > > To post to this group, send email to [email protected]. > > Visit this group at https://groups.google.com/group/link-grammar. > > For more options, visit https://groups.google.com/d/optout. > > > > -- > Ben Goertzel, PhD > http://goertzel.org > > "I am God! I am nothing, I'm play, I am freedom, I am life. I am the > boundary, I am the peak." -- Alexander Scriabin > > -- > You received this message because you are subscribed to the Google Groups > "link-grammar" group. > To unsubscribe from this group and stop receiving emails from it, send an > email to [email protected]. > To post to this group, send email to [email protected]. > Visit this group at https://groups.google.com/group/link-grammar. > For more options, visit https://groups.google.com/d/optout. > -- You received this message because you are subscribed to the Google Groups "opencog" group. To unsubscribe from this group and stop receiving emails from it, send an email to [email protected]. To post to this group, send email to [email protected]. Visit this group at https://groups.google.com/group/opencog. To view this discussion on the web visit https://groups.google.com/d/msgid/opencog/CAHrUA35kTedJyp_OJ6FwTGP5v-NvKx%2Bky%3DNmR_SuJ-ZzzT-VHQ%40mail.gmail.com. For more options, visit https://groups.google.com/d/optout.
