If this article is correct about limitations, couldn't one simply include a Turing machine model into the process to train algorithms?
Some ideas:
- The vectors are Turing tapes, or
- Each point in a tape is a DNN, or
- The "tape" is actually a "tree" each point in the tape
is actually a branch point of a tree with probabilities going each way, and the DNN model can "prune this tree" to refine the set of "spanning trees" / programs.
Or, hehe, maybe I'm leading people off track. I know absolutely nothing about DNN ( except I remember some classes on gradient descent and SVMs from bioinformatics ).
You can bolt all kinds of funny structures into some DNN system, but if the system doesn't have well behaved gradients (or if it isn't even differentiable) it won't train.
Yes, definitely. Pick up the early FRUiTS books by Phaidon for a great, enlivening and colorful overview of 90s street fashion and youth. These images of creativity are what first got me really interested in traveling to JP.
This probably has a big hole in it that I missed. But I just feel there must be some way to make a PK cryptosystem using Chinese Remainder Theorem. And I've found it in this one way at least.
I'm somewhat concerned this is simply the (already broken) Knapsack PK cryptosystem, but I don't know enough about that to say if it's different or not, essentially. Anyway, I know this community is not totally deep into reviewing crypto from random people, but I'm putting it out here as a record and in case anyone sees a hole the care to point out. Thanks!
Yeah, when you talk and track people's facial expressions in real time, you can see which particular things they disagree with, so you have useful feedback. A down vote is sort of like someone running up behind you and king hitting you, for something you said in some other group earlier, and maybe you don't know why. It's not always so non-specific...but not being able to respond to a downvote...makes it an incredibly potent tool to control discourse.
I think there might be a potential problem there in that retaliatory downvoting might be used as a weapon to disincetivise criticism, especially when (unavoidably) defensive retaliatory downvote cliques form.
So I think I would be okay with a mandatory anonymous reply as a precondition for downvoting.
Oh no I didn't mean it in a retaliatory sense specifically, I meant it like everyone having the option to do it, or no one except the few ( the mods? the ones who have objective standards ) having the options.
I meant like they can downvote me, and I can't downvote them, or anyone, at any time, because I don't have that whatever the requirement is ( mod level, karma level ) yet...so it's like this asymmetry that doesn't exist in the real world.
I don't feel any restriction talking to any human in real life. Nothing they can do to me I can't do back to them. So it's balanced. I'm safe. We're equal. I understand other people may feel differently, this is how I feel.
But on HN, I can't downvote. So there's this asymmetry. To me, it's a perversion. Because in the real world, only the State has a monopoly on legal violence. Only the state can use violence, legally. But on HN, downvotes don't only come from mods or police, they come from whoever has accrued sufficient karma -- and I don't equate this with a sort of objective, regulated, principle-based State level power.
So the state has a (correct, IMO) asymmetric monopoly on legal violence and can dish it out to people as it sees fit, but people cannot do the same, without either expecting to get the same in self-defense, or getting consequences from the state. But on HN, any random Jo with enough karma has the analogous asymmetry power.
That's what I meant. It's not necessarily about retaliating, it's about having that option on the table, rather than some people having it and others not having it.
Or no one having it except the State ( the mods, I guess ).
I think the downvoter should have to highlight the part(s) of the comment that caused them to downvote. This could provide useful signal to HNs moderators and moderation algorithms on the downvoted and the downvoters to help them calibrate the system. It could also be shown to the downvoted, or to everyone so it was clearer what the downvoting was targeted at.
It's a nice community-supporting sentiment but I think the idea that such a powerful tool as a down vote is only used for noble purposes, rather than to condemn and silence dissent or as an "easy out" substitute instead of a comment for something one disagrees with, is a fiction. Am I saying all downvotes are tools of censorship and punishment of diversity? No. But I feel down votes are powerful tools, and to me the idea that they are only used for good, while noble and community supporting, is naive.
So from the point of view that considers down votes something of a problem, what solution do I see? Nothing really conclusive. I think a step in the right direction is if downvotes only affected the comment they pertained to, and not overall reputation -- then I'd feel a lot more free.
Sure.....but then I think everyone ought to be able to down vote. Or only the mods. If it's a tool, it's either a tool for all, or a tool only wielded by the ones who define community standards, work to enforce them and act basically objectively in line with those principles anyway.
If it's not that objective tool, then it's a tool for everyone to wield subjectively. That's what I'd prefer.
I know others may feel differently. This is how I feel about it.
This page is an example why I gave up and still give up on advance mathematics.
Maths is just the extreme example of science, where to make communication possible between its practitioners, new words encoding known facts are constantly created. Then another layer of new words with definitions based on the first layers are defined... and so on. Rapidly, we end-up with total gibberish for the non-initiated.
For example on that page, we start with quintic and radicals (I can grok that). Then icosahedral functions, then hypergeometric functions, finite monodromy, 60-fold branched covering of the complex projective line (lost!).
This could have some holes, so I'm putting it out there to see if anyone can poke some in it, or seeing if it can stand up. It's flexible enough that it can encompass a family of ways of doing this -- but I'll consider it defeated if there's a fundamental flaw with the central idea of using AONTs to confound adversaries attacks. My plan with this is, if it does work out, not to patent it, but just to implement it securely under MIT license.
I currently consider this broken since, turns out that XOR and permutation do not commute over each other, and anyway to get them do so so requires knowledge of the secrets to be transmitted in a way to the channel that lets attacks recover them.
The need for XOR can be removed, but then each half of the AONT(message) is exposed. I couldn't see a way around this. So even tho I'm sure there is a way to construct a secret exchange on insecure channel mechanism (probably using 3 pass) I do not see it right now.
Some ideas:
- The vectors are Turing tapes, or
- Each point in a tape is a DNN, or
- The "tape" is actually a "tree" each point in the tape is actually a branch point of a tree with probabilities going each way, and the DNN model can "prune this tree" to refine the set of "spanning trees" / programs.
Or, hehe, maybe I'm leading people off track. I know absolutely nothing about DNN ( except I remember some classes on gradient descent and SVMs from bioinformatics ).