@ThomasKlimpel ? where did you see "withdrawing since this submission is in bad taste"?

@vzn Just click on v3 at the arXiv page, and look at the comments

oh ok thought there was maybe something wrong with my browser because the abs pg reloads, nothing else. ok, that comment only appears on v3 of the abstract pg. still have no idea why he would withdraw/ retract a careful 22pg analysis, makes no sense to me. would like to ask him what he means by that. also not sure what you mean by "maybe a sort of silent comment on subj of public amateur reviews". review does not seem "amateur"...

Yes, that was also one of my interpretations. The other interpretation is related to the acknowledgements:

> We thank professors Fabio L. Traversa and Massimiliano Di Ventra for helpful discussion and clarifications, and professors Hava T. Siegelmann, David A. Mix Barrington, and Scott Aaronson for their invaluable insights.

I still need to post the rest of my review. There is also a 3rd interpretation, at least with respect to F. Traversa and M. Di Ventra: For their earlier papers, they knew exactly that their proposed algorithms failed to be in one step (or better than known complexity bounds), they knew where they failed and intentionally introduced unsound arguments to still claim the contrary. For their digital memcomputers paper, at least they didn't knew whether their algorithm would fail.

Hence they probably defended their paper more vigorously, and also started a company to actually develop their algorithm into a product. And after their recent results, they probably actually start to believe that their newest algorithm might indeed be able to achieve what they claimed all along. Typical case of "fake it until you make it"...

@ThomasKlimpel was musing myself recently on how the startup angle complicates the science. dont think they mention their startup in all their papers, it may have been erected since the early ones. it decreases impartiality in some ways. have lots of personal experience in this area with the dotcom era (now ages ago). which reminds me, have you heard of the gartner hype curve? a nice encapsulation of some of these phenomena/ dynamics of the intersection of humans (psychology/sociology) and tech.

@ThomasKlimpel notable/ significant credits. iirc aaronson harshly criticized the memcomputing concept/ paper(s)...

Aaronson harshly criticized the "universal memcomputing machines" paper. He also mentions that apparently the authors knew all along what was wrong, but decided to only say it very hidden deep inside the paper. He explicitly did not appreciate that...

But Aaronson never reviewed the "digital memcomputers" paper. And I should say that there are also many positive things to be said about those papers. The authors really try to describe what they did, they include helpful pictures and details of simulations and experiments.

seems aaronson sometimes has a tendency to "throw out baby with bathwater" with his scathing reviews, there are multiple cases of this, Dwave comes to mind. but all seem to be "semi published" on his blog, not in papers... he is somewhat "careful" that way... agreed there is something worth looking into with memcomputing, think field is in very early stages & new insights theoretical + applied, maybe some quite substantial, are waiting to be uncovered. terra incognita

Saunders seems to provide no affiliation/ personal bkg in his papers, do you have any idea?

re aaronson dont know if you ever saw this, one of my earliest blog posts vzn1.wordpress.com/2013/02/20/…

He says: "I am a first-year M.S. student at the University of Massachusetts Amherst studying computer science." Sounds like a clear affiliation and bkg to me...

@ThomasKlimpel where? dont see it in paper.

"He says" is a link...

@ThomasKlimpel oh, ok, did you google him? maybe worth emailing him about all this.

from home pg, interesting blog! djsaunde.wordpress.com

lol he is still citing v2 of his paper on his home pg. so much for a retraction, a bit strange

Hacker News links to Scott Aaronson's inital reaction and interaction with Fabio: "Ah, so the authors do understand the problem. But why did they wait until page 11 to tell us this? Why didn’t they say so on page 1, so people could make an informed decision about whether to read further?" This is what I meant by "He explicitly did not appreciate that..."

oh correction it was v2 of the ventra paper he links to. arxiv.org/pdf/1211.4487v2.pdf but he still refs/ has a memcomputing paper review on his own web site. is this (nearly?) the same paper as the retracted one? djsaunde.github.io/survey-discussion-memcomputing.pdf

I will write to Daniel Saunders, but only after posting the rest of my review. But I will mostly tell him: "(1) nice work, (2) there is a way on arXiv to get the source of an article. Then you can at least quote from it without introducing unnecessary errors, and maybe even use pictures were appropriate without always needing to reference the original papers for the pictures. (3) Well, maybe directly reusing word feels less "illegal" than directly reusing pictures. Decide for yourself if its OK"

@ThomasKlimpel lets try to get/ persuade him on to stackexchange/ chat :) ... think all his ML+kaggle work looks very cool

Feel free to persuade him to join stachexchange / chat...

hes also has a math degree. btw cant remember what degree(s) do you have? djsaunde.wordpress.com/2017/01/11/…

I am a dipl(o)mat. ("Diplom Mathematiker").

its a club :) djsaunde.wordpress.com/2016/02/16/why-mathematics-matters

@ThomasKlimpel maybe beat you to it, he has comment moderation turned on, so no instant gratification for me, but just posted a comment on his memcomputing blog pg pointing in here/ inviting him djsaunde.wordpress.com/2016/11/15/…

The end of an error? Timothy Gowers on recent open reviews

> Fortunately, this can work both ways: the fact that the articles in the journal Homeopathy (formerly the British Homoeopathic Journal) have been through a process of formal peer review and are published by a major publisher does not lead to their being taken seriously by non-believers.

From foreXiv: For several months, Fermat’s Library has offered a Chrome extension called Librarian for browsing PDFs on the arXiv that automatically parses references to clickable journal links and bibtex entries.

Very recently they added the ability to publicly comment, visible to anyone else running Librarian. Should be lower friction than commenting on (also excellent) SciRate.

Back to the review of Daniel Saunders survey and discussion of memcomputing machines.

In section 4 "Memprocessors", the formal definition of a memprocessor from the "Universal Memcomputing Machines" paper is reproduced. One confusing part of that definition is the intended difference between x (the "state") and y (an "array of internal variables") of the memprocessor. The original paper also contains examples, but those don't resolve that question either. That section is not written in Danial Saunders words, but basically just quotes the relevant passages of the original paper.

In section 5 "Universal Memcomputing Machines", the definitions of universal memcomputing machines are reproduced, together with the properties that those are claimed to have. However, the concepts themselves are already sufficiently confusing, and they are just reproduced here without the original pictures to illustrate them, but instead with some unnecessary errors.

However, those inaccuracies forced me to reread the passages in the original "universal memcomputing machines" paper, and I finally understood the "UMM Information Overhead" concept. The overhead arises from the set of possible results which could be queried from the memcomputing machine. The concept is first illustrated at a simpler example, so the authors indeed try to convey something, and not just create confusion.

On the other hand, this overhead is so ephemeral that it is not sufficient for hoping that it can be exploited for any real exponential speedup.

Even so this section mostly just quotes passages of the original paper, some of the authors own words and opinions are intermixed, like: "This argument does not show that these machines are strictly more powerful than Turing machines, as was claimed." Apart from such obvious cases, the only way to distinguish between quoted passages and the authors own words in to compare with the original paper. This is not ideal.

In section 6 "UMM SUBSET-SUM implementation", the proposed memcomputing algorithm for solving the NP-complete subset-sum in polynomial time (or even a single step) is presented and discussed. Even so it still quotes large parts from the original paper, the presentation has been adapted by the author and is sufficiently under his control, so that he doesn't introduce unnecessary errors like in the previous section.

At this point I downloaded the latex sources of both the survey and the "universal memcomputing paper" and compared whether the survey reused those. It did not, the latex was completely different.

The objections of Scott Aaronson against this proposed algorithm are repeated and confirmed to be valid. I already mentioned earlier that for me, the narrow bandpass filter operating in no-time is the most glaring and obvious slight of hand. In a sense, the author is happy that Scott Aaronson already took the trouble to refute the algorithm, so that he does not need to defend an argument against it on his own.

In section 7 "Digital Memcomputing Machines", relevant definitions and concepts from the "digital memcomputers" paper are reproduced. Even so one still has to read parts of the original paper, the compression to a few pages at least motivates to work through the corresponding passages in the original paper. The main contribution of the author here is the selection of the material, no obvious judgements seem to be contained in this section.

In section 8 "DMM Implementation", the concepts of self-organizing logic gates are reproduced. The section is short, but equation (21) and (22) contradict each other, and the descriptions are to short to understand what is going on. Reading the material in the original paper reveals how equation (21) and (22) should have been written, and that all the important details are actually described. The original paper is impressive at this point, by also describing possible physical implementations.