2018-05-24 8:42 GMT+03:00 Linas Vepstas <[email protected]>: > > Actually, I want you to not think about this. I strongly believe that > pretty much anything you can think of will fit nicely into an Atom, or into > a Value. I do not want to see a third kind of "generic object system" > being created, that would be a deep mistake. >
OK. Actually, I was not too interested in something between Values and Atoms, but more in something between deep learning and opencog. The latter doesn't necessarily requires the former. So, I promise not to think about this, unless find it necessary :) Yes, here: https://github.com/opencog/atomspace/blob/master/opencog/ > sheaf/docs/sheaves.pdf > Thanks. I think the newcomers need to have another half-year-ish of hands-on > experience before we debate such fairly significant architectural changes. > I understand your worries, but - we don't debate right now, but ask. - we have tasks, we would like to start solving right now (of course, it's up to Ben to tell us to gain another half-year-ish of hands-on experience on secondary issues). Nothing I've heard so far requires any changes at all, and I can see a > reasonable, simple solution, just fine. > I'm not sure what do you mean by "any changes at all", while even basic problems require some changes (e.g. TensorValue, TimesLinkValue, etc.). And that is not what I was saying, at all. What I was talking about was the > principles of software architecture > I was talking not about software development, but about R&D. Again, it would be great if we could nail down the next level of details. > Sure, sure. Actually, SynerGANs and VQA example are already enough for the discussion, but I guess I need to describe them to you in more detail. I hope to return to this in 10 days with more details and more examples. Let's take a small break until then. > This would allow us to write tensorflow programs in Atomese. Why is that > interesting? Not because we want humans to write tensorflow models in > atomese, but because maybe we can have PLN perform reasoning about tensorflow > models, or because we can use MOSES to create, control and evaluate > tensorflow models, or perhaps you have so probbilistic-programing idea that > could auto-general different tensorflow models. > > Yes. Automatic design of DNNs is another example I have in mind. Didn't I mention it? Nevermind. I'm afraid I need to write a very long document to describe all aspects. I just tried to focus on the first small step regarding applying PM to Values. But it seems we need the whole picture to proceed. -- Alexey -- 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/CABpRrhwey9Jy7n8qExYXa_cqPojzKjf7RgPA2vZN2b6A6hKt%2Bg%40mail.gmail.com. For more options, visit https://groups.google.com/d/optout.
