Supermathematics and Artificial General Intelligence

[I_am_not_mafia]

Post: [FoR&AI] The Seven Deadly Sins of Predicting the Future of AI

Haven't read this yet but a quick skim through looks good. Rodney Brookes is a big name in AI. I loved his idea for a subsumption architecture.

[causal code]

I always wondered how they would solve reinforcement learning time-dependency problems, because RL had some issues with time dependent games.

But, it looks like "manifold learning" allows for deep learning models to "generalize across time", which is cool.

Some interesting papers:

"VAE" networks, for computation wrt to features like position across time:
https://arxiv.org/abs/1606.05579

Improving the long term temporal dependency of recurrent networks:
https://deepmind.com/blog/decoupled-neur...gradients/

[causal code]

It looks like for months now, the op has made progress on the thing he described.

I had given up on trying to find his or her paper a month ago, (because there are no links in the op), but I think I'd found his or her paper on quora some time after I had given up, and it's in this link.

In the paper, he or she mentions a "transverse field ising super hamiltonian", in relation to another paper.

Here is a screenshot of the experiment designed in the paper (downloaded from the project's github page, linked in the paper):



Can anybody make sense of the above?

I also notice that the paper which he or she bases everything on called "Supersymmetric methods... at brain scale" was not cited by other research groups, although it's on arxiv. Is this a safe paper to bet on?

[I_am_not_mafia]

Isn't the OP is ProgrammingGodJordan who has created many different sock accounts in the past? The format is the same. If it is the same person then I wouldn't expect any published results because they do not know how to write a paper.

[LastPoet]

Isn't the OP is ProgrammingGodJordan who has created many different sock accounts in the past? The format is the same. If it is the same person then I wouldn't expect any published results because they do not know how to write a paper.

It is something betwixt that.

[causal code]

Isn't the OP is ProgrammingGodJordan who has created many different sock accounts in the past? The format is the same. If it is the same person then I wouldn't expect any published results because they do not know how to write a paper.

Apart from the light blue colours, the paper looks pretty much normal, with strong references, many from arXiv.

The only probably worrisome exception I can detect so far, is the citation of the Supersymmetric methods paper by Perez et al., which is on arxiv, and looks very complicated, but has not been cited by many people.

I am wondering if the fact that the paper is on arxiv has much credence to the OP citing the paper.

[I_am_not_mafia]

He loses credence with the opening paragraphs in his github project and the downloadable article  ...

This thread concerns attempts to construct artificial general intelligence, which I often underline may likely be mankind’s last invention.

I clearly unravel how I came to invent the supermanifold hypothesis in deep learning,

First rule of science is to not claim anything more than what can be backed up by the results otherwise your paper will get rejected and your research will never be built upon by someone else. Which is preferable to being crucified when you're standing up there presenting your work.

[causal code]

It looks like for months now, the op has made progress on the thing he described.

I had given up on trying to find his or her paper a month ago, (because there are no links in the op), but I think I'd found his or her paper on quora some time after I had given up, and it's in this link.

In the paper, he or she mentions a "transverse field ising super hamiltonian", in relation to another paper.

Here is a screenshot of the experiment designed in the paper (downloaded from the project's github page, linked in the paper):



Can anybody make sense of the above?

I also notice that the paper which he or she bases everything on called "Supersymmetric methods... at brain scale" was not cited by other research groups, although it's on arxiv. Is this a safe paper to bet on?

Ooops, I accidentally deleted the screenshot from my tinypic account above. Deleted the wrong duplicate.

Here is the screenshot of the OP's somewhat gibberish paper (could be because I didn't go past Calculus I?):



I have a question to ask you Mathilda about reply #27. Going to get some food and return.

[causal code]

He loses credence with the opening paragraphs in his github project and the downloadable article  ...

This thread concerns attempts to construct artificial general intelligence, which I often underline may likely be mankind’s last invention.

I clearly unravel how I came to invent the supermanifold hypothesis in deep learning,

First rule of science is to not claim anything more than what can be backed up by the results otherwise your paper will get rejected and your research will never be built upon by someone else. Which is preferable to being crucified when you're standing up there presenting your work.

You're right, I don't see any results in the OP's paper, and that's a really bad sign.

My 10 cents:

According to a quora answer, the OP's paper seems to be talking about turning the "Hamiltonian" seen in quantum reinforcement learning into a "Super Hamiltonian" instead, inspired by "evidence" about supersymmetry in the brain. 

(If Perez et al.'s paper about supersymmetric "evidence" in biological brain is valid, then the OP's paper would be interesting, because it says this would be a way to "capture" certain "physics priors", and in reality, babies are actually observed to know physics intuitively according to Stahl et al)

Eta: The "Transverse Field Ising Spin Super Hamiltonian" in the screenshot from reply 28, seems to be possible, according to this "SQCD" paper.


I also notice the OP's paper cites yet another paper, that also has no experimental results. "The Consciousness Prior" by Yoshua Bengio.

Anyway, forgetting about the OP's paper for now, what are your thoughts on Bengio's paper Mathilda?