ERDNASE

Discuss general aspects of Genii.
Carlo Morpurgo
Posts: 375
Joined: January 17th, 2008, 12:00 pm
Location: Columbia, MO

Re: ERDNASE

Postby Carlo Morpurgo » May 3rd, 2018, 9:01 pm

I recommend browsing the first several (even just 5) pages of this thread, from way back (2003). I just did, and it's pretty amazing how different things were back then. Stimulating, fun, civilized. I do not find this stimulating at all, maybe because the tone of the discussion is just plain annoying. But maybe something will come out of it, who knows.

Leonard Hevia
Posts: 1951
Joined: January 17th, 2008, 12:00 pm
Favorite Magician: Dai Vernon, Frank Garcia, Slydini, Houdini,
Location: Gaithersburg, Md.

Re: ERDNASE

Postby Leonard Hevia » May 3rd, 2018, 9:31 pm

Carlo Morpurgo wrote:I recommend browsing the first several (even just 5) pages of this thread, from way back (2003). I just did, and it's pretty amazing how different things were back then. Stimulating, fun, civilized. I do not find this stimulating at all, maybe because the tone of the discussion is just plain annoying. But maybe something will come out of it, who knows.


Something did come out recently, Bob Coyne's updated essay comparing Sanders' writing with Erdnase:

http://www.cs.columbia.edu/~coyne/erdna ... guage.html

You're going to have to cut thru the noise such as Keyes' buffoonery and Wasshuber's Gallaway infomercials.

User avatar
lybrary
Posts: 1169
Joined: March 31st, 2013, 4:59 pm
Contact:

Re: ERDNASE

Postby lybrary » May 3rd, 2018, 9:39 pm

Brad Henderson wrote:perspective shifts
look it up on the book you have but whose information you want to ignore.
Joe Crist writes "perception distortion". Don't know if that is what you meant. If you did, then going from "perception distortion" to "perspective shifts" is quite an accomplishment, Brad. As I said, you don't know what you are talking about, but you still like to talk about it. That is not what you call 'objective', that is misinformed. But perhaps that is what you meant - misinformed. It is hard to know. You make errors in every second post. I guess too many flies in your room.

Brad Henderson wrote:drawing red lines is not proof of facial recognition
Another subject you know nothing about. Facial recognition uses spatial distances of various facial features. Drawing red lines to compare where various facial features are located is therefore a form of facial recognition. Just because you don't understand it, doesn't mean it isn't.

Brad Henderson wrote:owning a book your company printed isn’t proof of card handling skills
That is true, and I have never argued that this is proof of skill of sleight-of-hand. But it is very likely proof of interest, particularly since it is supported by his ownership of gambling books, by his use of magic phrases in places where one does not expect them, by his titling of his theater number "The Magic Wand". All of that supports the argument that he had this book because of its contents and not solely because it was printed where he worked. Additionally, we would expect that the author would keep a copy of the book in his library. That puts Gallaway in a group of people who have a much higher likelihood of being Erdnase, than any other random guy plugged from the street.

Brad Henderson wrote:and saying you did not recognize a drawing is not the same as saying there were photographs.
But that was Smith's own reasoning. He did not recognize the illustrations and therefore he thought he must have traced them from photographs. Neither Gardner nor Smith are alive, so we will never know what was indeed meant by that statement. All we have are the words, and the words say exactly that. He thought he probably did them from photographs.
Lybrary.com Magic & Gambling
preserving magic one book at a time

Brad Henderson
Posts: 4546
Joined: January 17th, 2008, 12:00 pm
Location: austin, tx

Re: ERDNASE

Postby Brad Henderson » May 3rd, 2018, 10:38 pm


Joe Crist writes "perception distortion". Don't know if that is what you meant. If you did, then going from "perception distortion" to "perspective shifts" is quite an accomplishment,


you knew exactly what i meant. but you keep focusing on the fly [censored]

since you can’t make your case, you pick up on typos as if that proves anything.

you’re a dishonest actor.

and no,
your red lines don’t prove anything.

and would an author have an incriminating copy of his book? as opposed to selling them for the money?

your position is built on assumptions.

you’re just mad that we won’t drink the kool aid and now you’re lashing out.

shift - distortion - all means they weren’t done from photographs.

three strikes chris. you’re out.

hey i just used a sports metaphor. must be proof that i’m a tremendous athlete.

Bob Coyne
Posts: 717
Joined: January 26th, 2008, 12:00 pm
Favorite Magician: Charlies [sic]
Location: New York, NY

Re: ERDNASE

Postby Bob Coyne » May 3rd, 2018, 11:05 pm

Carlo Morpurgo wrote:I recommend browsing the first several (even just 5) pages of this thread, from way back (2003). I just did, and it's pretty amazing how different things were back then. Stimulating, fun, civilized. I do not find this stimulating at all, maybe because the tone of the discussion is just plain annoying. But maybe something will come out of it, who knows.

Yes, so true....Unfortunately it often feels like a battle zone now! It's a stark contrast how much more agreeable and constructive the tone was back then. It's good to be reminded of that. Hopefully some of that positive spirit can return.

It's funny too, since some of the same topics (e.g. tracing vs drawing the illustrations) were being discussed. But somehow it didn't become dogmatic, even if there were differing opinions.

User avatar
lybrary
Posts: 1169
Joined: March 31st, 2013, 4:59 pm
Contact:

Re: ERDNASE

Postby lybrary » May 3rd, 2018, 11:26 pm

Brad Henderson wrote:hey i just used a sports metaphor. must be proof that i’m a tremendous athlete.
Not proof that you are a tremendous athlete, but you are probably interested in sports to some degree. But sports is a bad analogy because it is too prevalent. Sports metaphors have become ubiquitous and therefore don't mean much in terms of the author's background. But magic is not something that is or was that ubiquitous. It is therefore a very interesting signature if an author uses many magic phrases in contexts where one would not expect them. Please show me an author of non-fiction books/articles, who hasn't written any magic books, and who is as far as one can tell not interested in magic, who in several consecutive publications consistently uses magic phrases. In Gallaway's case he had four consecutive non-magic publications where he uses magic phrases: The Monotype System, Copyfitting, Printing Estimating/Printing Practice, Estimating for Printers. That is not only a revealing linguistic fingerprint in and of itself, but also suggest a certain affinity to magic, which is supported by his ownership of a copy of EATCT, as well as the title he chose for his performance: "The Magic Wand". Denying these facts is the only dishonesty you should be worried about.
Lybrary.com Magic & Gambling
preserving magic one book at a time

Roger M.
Posts: 1598
Joined: January 17th, 2008, 12:00 pm

Re: ERDNASE

Postby Roger M. » May 4th, 2018, 1:10 am

The thread changed tone entirely when Chris began posting here, and decided to manipulate it such that he could not only become revered as the "discoverer of Erdnase", but that he could further abuse the thread and subject matter in a blatant effort to advertise and market his Lybrary wares at no cost to him.

The thread will remain this way as long as Chris is posting here.

The danger is always that, left to his own fabrications, and without the ongoing point/counterpoint ... the record begins to let Chris do what he's been here to do from the very beginning ... which is to falsely claim ownership of title - "Discoverer of Erdnase".
Considering the importance of this thread in the broader search for Erdnase, that would be an incredible shame.

Further, when you add guys who wander from Genii thread to Genii thread seeking a confrontation, who post endlessly without saying anything, and who have absolutely no interest in the subject matter ... who are just desperately seeking an anonymous internet argument - and you have this thread in May 2018.

It's a long thread, and it will no doubt go through a many more iterations in its lifetime ... hopefully sans the delusional fabricators.

User avatar
jkeyes1000
Posts: 483
Joined: August 2nd, 2012, 3:12 pm

Re: ERDNASE

Postby jkeyes1000 » May 4th, 2018, 5:23 am

Brad Henderson wrote:

Joe Crist writes "perception distortion". Don't know if that is what you meant. If you did, then going from "perception distortion" to "perspective shifts" is quite an accomplishment,


you knew exactly what i meant. but you keep focusing on the fly [censored]

since you can’t make your case, you pick up on typos as if that proves anything.

you’re a dishonest actor.

and no,
your red lines don’t prove anything.

and would an author have an incriminating copy of his book? as opposed to selling them for the money?

your position is built on assumptions.

you’re just mad that we won’t drink the kool aid and now you’re lashing out.

shift - distortion - all means they weren’t done from photographs.

three strikes chris. you’re out.

hey i just used a sports metaphor. must be proof that i’m a tremendous athlete.


The trouble with this contention (traced vs. not traced) is that neither adversary wishes to concede the possibility that some were, and some were not.

Smith looked at the illustrations and "couldn't remember' having drawn all of them. Maybe he didn't.

Others have examined them and concluded that they must have been traced. But how thoroughly have they scrutinised each one?

In order to settle this question, you need to compare each to each, and never assume that they were all done the same way.

To me, it is clear that some were traced (as Smith suggested), and some were drawn rather sloppily with the free hand of another 'artist' (which might explain why Smith didn't recognise them).

Brad Henderson
Posts: 4546
Joined: January 17th, 2008, 12:00 pm
Location: austin, tx

Re: ERDNASE

Postby Brad Henderson » May 4th, 2018, 8:03 am

lybrary wrote:
Brad Henderson wrote:hey i just used a sports metaphor. must be proof that i’m a tremendous athlete.
Not proof that you are a tremendous athlete, but you are probably interested in sports to some degree. But sports is a bad analogy because it is too prevalent. Sports metaphors have become ubiquitous and therefore don't mean much in terms of the author's background. But magic is not something that is or was that ubiquitous. It is therefore a very interesting signature if an author uses many magic phrases in contexts where one would not expect them. Please show me an author of non-fiction books/articles, who hasn't written any magic books, and who is as far as one can tell not interested in magic, who in several consecutive publications consistently uses magic phrases. In Gallaway's case he had four consecutive non-magic publications where he uses magic phrases: The Monotype System, Copyfitting, Printing Estimating/Printing Practice, Estimating for Printers. That is not only a revealing linguistic fingerprint in and of itself, but also suggest a certain affinity to magic, which is supported by his ownership of a copy of EATCT, as well as the title he chose for his performance: "The Magic Wand". Denying these facts is the only dishonesty you should be worried about.


no chris. i have no interest at all in sports.

never have.

but it’s good to see you are always shifting goal posts.

see what i did there.

so using sports references is different than magic ones.

got it.

so convenient for you.

i suppose all those political cartoonists who used the vanishing lady in their work were also interested in magic - to the degree that erdnase had to be

see that’s the problem. We aren’t dealing with someone who had a casual interest. We are dealing with someone who was one of the best that ever were.

and you want me to believe he held a steady job and had a wife and kids?

you don’t know what you are talking about.

of course your buddy keyes has told us at this time everyone wanted to be a magician and loved magic, so the use of these terms wouldn’t have been uncommon at all.

or is he wrong?

User avatar
jkeyes1000
Posts: 483
Joined: August 2nd, 2012, 3:12 pm

Re: ERDNASE

Postby jkeyes1000 » May 4th, 2018, 8:33 am

Brad: I did not say that everyone wanted to be a magician in the early 1900s. I simply pointed out that it was amongst the most common aspirations for a young boy. Others being soldiers, sailors, firemen, train conductors, "big game" hunters, etc.

Brad Henderson
Posts: 4546
Joined: January 17th, 2008, 12:00 pm
Location: austin, tx

Re: ERDNASE

Postby Brad Henderson » May 4th, 2018, 9:57 am

ah. still quibbling over the fly [censored] i see

chris still is in trouble. If interest in being a magician were as common as you suggest, then the casual use of superficial magical terms would be irrelevant to his case.

so which is it chris? is keyes correct? or are you wrong?

Bill Mullins
Posts: 5911
Joined: January 17th, 2008, 12:00 pm
Location: Huntsville, AL
Contact:

Re: ERDNASE

Postby Bill Mullins » May 4th, 2018, 1:29 pm

lybrary wrote:
Brad Henderson wrote:drawing red lines is not proof of facial recognition
Another subject you know nothing about. Facial recognition uses spatial distances of various facial features. Drawing red lines to compare where various facial features are located is therefore a form of facial recognition. Just because you don't understand it, doesn't mean it isn't.

Chris has referred to his exercise in drawing red lines several times as "facial recognition" (I'll call it FR), and asserted that he has identified the unknown Monotyper as being Gallaway.

I spent some time reviewing several overviews of FR technology. In no way is what Chris has done close to how FR works, and he's fooled himself if he believes what he has done would come close to yielding an accurate result. FR is a very difficult problem, and getting an answer is far more complex than saying you've matched the relative heights of several facial features of two people.

The first part of the problem is getting a useful image of the face to identify. One overview says " the face needs to be normalized. This means that the image must be standardized in terms of size, pose, illumination, etc., relative to the images in the gallery or reference database . . . it is essential that the probe [face you are trying to identify] is as close as possible to a standardized face." Another says, "The initial and all subsequent biometric acquisitions must meet the same image quality standards." This is why passport photos are standardized. If you have gotten a new drivers license photo lately and been told not to smile, this is why. It is why, when you travel internationally and a machine takes your photo to match it to your passport, you have to stand in a particular place relative to the camera, and the lighting is particular, and you must stare directly at the camera, and you may have to do it more than once to get a good photo.

While the "gallery" image (the identified face to be matched against) of Gallaway is pretty good, the "probe" image is far from ideal. It is of a head rotated away from the camera, perhaps by as much as 120 degrees. Only one ear is visible. No details of the eyes or eyebrows or mouth are visible. The nose is seen only in partial profile. You cannot ascertain the width of any facial features, only their heights. The illumination is different, and the face is in shadow. It is at a different scale, and was blown up such that resolution is poor. In fact, given that the face is the front of the head, and we are seeing the back right aspect of the head, it is arguable that this is even a "face" at all. It need hardly be pointed out that the image is in no way "normalized", and is in no way is sufficiently clear or detailed to recognize.

The importance of a common standard format for normalization is such that the ISO has developed a appropriate standard.

There is data that shows how some of these problems affects accuracy. One study found that when otherwise good quality photo pairs (same subject, same camera setup, same illumination, same resolution) included one frontal view and on ~60 degree profile view, accuracy (for three different algorithms under test) dropped from (~98%, ~96%, ~85%) to (~29%, ~63%, ~44%) (compare fig 7 to fig 9). It also showed drops in accuracy when illumination was changed, and when scale/resolution was changed. Note that this only changes one variable at a time, and used much more sophisticated matching strategies than Chris did. He is attempting to match two images which are non-normalized in every possible dimension, and he is using very rudimentary means of comparison.

Further, Chris hasn't pulled enough data from them to show a match. Real-world algorithms measure dozens of data points from photos to compare to a known example. They typically compare dozens of "nodal points" (absolute and relative locations of various features, such as the edge of the nostrils, the centers of the pupils, corners of the eyes, corners of the lips, height of the upper and lower lips, size and shape of the eyebrows, hairline, etc.) from two different photos to establish a similarity. Obviously, for the photo of the Monotype Typesetter, these features cannot be located. The relative heights of a half-dozen features is simply not enough data to do the job.

The two images which Chris has claimed that they "match" cannot be said to do so. They are far too dissimilar in format, pose, illumination, and resolution to extract the necessary data to demonstrate that they are of the same person. The more he comes up with goofy stuff like this to "prove" his case, the more he damages it.

User avatar
jkeyes1000
Posts: 483
Joined: August 2nd, 2012, 3:12 pm

Re: ERDNASE

Postby jkeyes1000 » May 4th, 2018, 1:33 pm

Apparently, Brad thanks that I am allied with Chris, and he is trying to "divide and conquer" the anti-Sanders forces.

That's not the way to win, Brad. You need to be smarter than the opposition, not more devious.

Good luck with that.

Jonathan Townsend
Posts: 8704
Joined: January 17th, 2008, 12:00 pm
Location: Westchester, NY
Contact:

Re: ERDNASE

Postby Jonathan Townsend » May 4th, 2018, 2:07 pm

Carlo Morpurgo wrote:I recommend browsing the first several (even just 5) pages of this thread, from way back (2003). I just did, and it's pretty amazing how different things were back then. Stimulating, fun, civilized. ...


+1

This kind of puzzle solving doesn't need to get personal or political. We've got a printer, a text, and some arguments about authorship. The Alexander/Hatch/Morpugo et. al. background data/text discussions are interesting too. Thanks guys.
Mundus vult decipi -per Caleb Carr's story Killing Time

User avatar
lybrary
Posts: 1169
Joined: March 31st, 2013, 4:59 pm
Contact:

Re: ERDNASE

Postby lybrary » May 4th, 2018, 5:00 pm

The facial comparison I did is much more detailed than all the characteristics Smith mentioned about Erdnase combined. Candidates have been eliminated simply because they were too tall. That is only one measure and it is imprecise because we only have Smith's recollections to compare against. Nevertheless, it is data that is useful.

Here we have two photos from different angles, not ideal, but they do provide a lot of details about the persons depicted. It allows a direct comparison of several features of the face. How far the nose is from the mouth, how the ear is positioned in relation to the eyes, nose, and mouth, how far the eye brows are away from the top of the head, etc. Having these things match is confirming evidence that the person shown could very well be Gallaway. If one looks at the control image I included, then one can clearly say that this is not Gallaway. Most other heads would not match all the distances identified by the red lines, and could therefore be shown to not be Gallaway. I am not saying this is a unique identification, but it is very strong confirmation that the author could indeed be Gallaway. If it would not be Gallaway, the likelihood that some feature would not match is very high.
Lybrary.com Magic & Gambling
preserving magic one book at a time

Bob Coyne
Posts: 717
Joined: January 26th, 2008, 12:00 pm
Favorite Magician: Charlies [sic]
Location: New York, NY

Re: ERDNASE

Postby Bob Coyne » May 4th, 2018, 5:58 pm

chris, What sort of validation has this technique undergone? If it's being done just to show the person in the photo could be gallaway, i think it achieves that. But to show that it *probably* is gallaway would requite some rigorous testing on the technique itself. It would then be possible to give some probability. i.e if you're trying to go beyond the limits of our intuitive judgments, then the technique's accuracy must be tested and quantified.

User avatar
lybrary
Posts: 1169
Joined: March 31st, 2013, 4:59 pm
Contact:

Re: ERDNASE

Postby lybrary » May 4th, 2018, 8:30 pm

That is funny, Bob, that you from all the people here are the first who requires quantification. You have argued vehemently against quantification of your own list of text examples. It is your double standard on display for everybody to see. Once you provide rigorous quantification and validation of your linguistic case for Sanders I will provide validation numbers for the facial recognition I have done. But let me say that facial recognition from a profile rather than the front, even one from bad angles, is being researched and the results are astonishingly good.
Lybrary.com Magic & Gambling
preserving magic one book at a time

Bob Coyne
Posts: 717
Joined: January 26th, 2008, 12:00 pm
Favorite Magician: Charlies [sic]
Location: New York, NY

Re: ERDNASE

Postby Bob Coyne » May 4th, 2018, 8:41 pm

lybrary wrote:That is funny, Bob, that you from all the people here are the first who requires quantification. You have argued vehemently against quantification of your own list of text examples. Once you provide rigorous quantification and validation of your linguistic case for Sanders I will provide validation numbers for the facial recognition I have done. But let me say that facial recognition from a profile rather than the front, even one from bad angles, is being researched and the results are astonishingly good.

I think you're missing my point. I'm saying that when you can't use normal intuition and instead want to rely on some quantitative technique, then that technique itself needs to be validated. There's nothing inconsistent about that.

I don't know if your technique is valid or not. I'm only saying that since normal intuition/perception doesn't apply (at least it doesn't for me -- i can't tell from that photo if it's gallaway or not), then it's not enough just to say these measurements show such and such without knowing how that metric has been tested more generally and with what sort of confidence level it can make predictions. This is standard operating procedure. Perhaps you've done that. I'm just asking.

Bill Mullins
Posts: 5911
Joined: January 17th, 2008, 12:00 pm
Location: Huntsville, AL
Contact:

Re: ERDNASE

Postby Bill Mullins » May 4th, 2018, 8:51 pm

Bob -- that's Chris's subtle way of saying, "No, I don't have validation." He doesn't validate his linguistic techniques, either. He doesn't validate his "analysis" of finger shapes, he doesn't offer any explanation other than his say-so for most of what he calls evidence. For Chris, "you can't prove me wrong" is the functional equivalent of "really strong evidence for".

For numerical validation, look at my post of earlier today: using a face image only 60 degrees off straight-on and holding all other variables constant reduces the ability to match faces significantly. Chris is using a face that appears to be 120 or more degrees off straight-on, and degrades several other variable as well. He claims matches on the eye and the mouth, but realistically, you can't even locate either one on the probe face -- either one could be a few millimeters above or below where he shows the red line.

User avatar
lybrary
Posts: 1169
Joined: March 31st, 2013, 4:59 pm
Contact:

Re: ERDNASE

Postby lybrary » May 4th, 2018, 9:02 pm

Bob Coyne wrote:I'm saying that when you can't use normal intuition ...
I don't know what you mean with 'normal intuition'. Is that different to just 'intuition'? What about common sense? Or what about two photos where it is for a six year old visible that all the features line up? My photo comparison requires no intuition. The features line up. Other features match, too. Light complexion. Bald or thinning hair on the top of his head. Not 100% proof that this is the same person, but since all the features one can compare match, the likelihood for them being one and the same is very high. That is much more intuitive than a list of phrases such as 'too good' that your list claims is proof of authorship identity.
Lybrary.com Magic & Gambling
preserving magic one book at a time

Jackpot
Posts: 236
Joined: June 8th, 2016, 12:38 am
Location: Santa Rosa, CA

Re: ERDNASE

Postby Jackpot » May 4th, 2018, 9:25 pm

lybrary wrote:Or what about two photos where it is for a six year old visible that all the features line up?

I hope this is a figure of speech. I have literally -- not figuratively -- worked with over a 1,000 six year olds. Many things which are "visible" to a six year old are not real. Relying on the judgement of six year olds only weakens your argument. Now if you have someone skilled in facial recognition finding your two red lined photos obvious you're on to something.
Not the one who created the Potter Index.

User avatar
lybrary
Posts: 1169
Joined: March 31st, 2013, 4:59 pm
Contact:

Re: ERDNASE

Postby lybrary » May 4th, 2018, 9:39 pm

Bill Mullins wrote:For numerical validation, look at my post of earlier today: using a face image only 60 degrees off straight-on and holding all other variables constant reduces the ability to match faces significantly.
That is your petty attempt to discredit face recognition. Of course, using a system that has been trained and optimized for frontal face recognition for a profile case will not produce great results. But a system trained for that particular case works a lot better.

Since Bill's pseudo-science knowledge is so dishonest I suggest for anybody who is genuinely interested in this subject to read for example this article http://www.rci.rutgers.edu/~vmp93/Confe ... 16_CFP.pdf
What you will learn is that recognizing profile images from frontal images is a common problem in surveillance and other real world applications. You will also learn that while Frontal-Frontal accuracy of face recognition algorithms is above 95%, the currently best Frontal-Profile algorithms achieve 85% accuracy. That is merely a 10% loss, which given the problem, is quite remarkable. So yes, one can recognize a person with decent accuracy from the profile without knowing the profile in advance.
Lybrary.com Magic & Gambling
preserving magic one book at a time

Bob Coyne
Posts: 717
Joined: January 26th, 2008, 12:00 pm
Favorite Magician: Charlies [sic]
Location: New York, NY

Re: ERDNASE

Postby Bob Coyne » May 4th, 2018, 9:58 pm

lybrary wrote:
Bob Coyne wrote:I'm saying that when you can't use normal intuition ...
I don't know what you mean with 'normal intuition'. Is that different to just 'intuition'? What about common sense? Or what about two photos where it is for a six year old visible that all the features line up? My photo comparison requires no intuition. The features line up. Other features match, too. Light complexion. Bald or thinning hair on the top of his head. Not 100% proof that this is the same person, but since all the features one can compare match, the likelihood for them being one and the same is very high. That is much more intuitive than a list of phrases such as 'too good' that your list claims is proof of authorship identity.

It's the same. And good old fashioned common sense is the same too. Basically what we can tell using our normal faculties. And normal perception is related. I really don't want to get into a semantic debate about the terminology.

However, when you start relying on tables of numbers and quantitative metrics, you're moving further from from intuition (normal or otherwise). It becomes harder and harder to know what they mean and what they imply and whether they can be trusted and for what exactly. So some sort of validation is needed to circumscribe all of the above, as well as the probabilities and error bars. When a physicist or engineer calculates forces and predicts that with this much fuel and these external forces (gravity, friction, etc) the space craft will do such and such, then our normal intuitions don't apply and we need to rely on validated techniques.

The probe photo of gallaway is from the back and I can't directly perceive if it's the same person or not. Nor do I get any vaguer intuitive sense that it's the same person (based on posture or whatever). All I can perceive or intuit is that it could be the same person...i.e. they're compatible. To go beyond that, I would then have to trust in your technique and whatever assumptions you're basing it on and whatever error bars it implies. But as far as I can tell it hasn't been validated and is based on shaky assumptions (as Bill pointed out).

My linguistic correspondences are something completely different. First they're intended as a way of directing our normal intuitions (yes, that again!) to salient examples. As I said at the beginning of all this, by pointing at something of significance, you can make it much more clear and understand it better. For example, when Carlo pointed to this thread back in 2003, it was obvious to anyone who took a look that things have changed (and how they've changed). Secondly, "too good" isn't intended as just a matching linguistic phrase. That's a minor part since it's polysemous and takes on additional meaning in context. Instead it's the phrase plus the particular sense and underlying psychology that's very revealing. And that's done in the larger context of approximately 140 other examples. Not to mention other types of linguistic evidence (use of colloquialisms, parenthetical question marks, biographical indices, etc).
Last edited by Bob Coyne on May 4th, 2018, 10:01 pm, edited 2 times in total.

Jackpot
Posts: 236
Joined: June 8th, 2016, 12:38 am
Location: Santa Rosa, CA

Re: ERDNASE

Postby Jackpot » May 4th, 2018, 9:59 pm

lybrary wrote:http://www.rci.rutgers.edu/~vmp93/Conference_pub/WACV2016_CFP.pdf

I'm not sure this article applies to your argument. This is about the comparisons of the front of the face to it's profile, but your examples compare the face to "mostly the back of the head". The article also relied on the pictures of celebrities for whom it would be easy to find perfect examples. If we know anything about Erdnase it was that he was not a celebrity.
Not the one who created the Potter Index.

Bill Mullins
Posts: 5911
Joined: January 17th, 2008, 12:00 pm
Location: Huntsville, AL
Contact:

Re: ERDNASE

Postby Bill Mullins » May 5th, 2018, 12:47 am

lybrary wrote:Since Bill's pseudo-science knowledge is so dishonest I suggest for anybody who is genuinely interested in this subject to read for example this article http://www.rci.rutgers.edu/~vmp93/Confe ... 16_CFP.pdf
What you will learn is that recognizing profile images from frontal images is a common problem in surveillance and other real world applications. You will also learn that while Frontal-Frontal accuracy of face recognition algorithms is above 95%, the currently best Frontal-Profile algorithms achieve 85% accuracy. That is merely a 10% loss, which given the problem, is quite remarkable. So yes, one can recognize a person with decent accuracy from the profile without knowing the profile in advance.


True, FR algorithms have advanced, and some do quite well on well-behaved frontal/profile data sets. So take the algorithms that they used, and apply them to your pair of images. Report that data.

Jack Shalom
Posts: 1368
Joined: February 7th, 2008, 12:00 pm
Location: Brooklyn NY

Re: ERDNASE

Postby Jack Shalom » May 5th, 2018, 10:17 am

lybrary wrote:
Since Bill's pseudo-science knowledge is so dishonest I suggest for anybody who is genuinely interested in this subject to read for example this article http://www.rci.rutgers.edu/~vmp93/Confe ... 16_CFP.pdf
What you will learn is that recognizing profile images from frontal images is a common problem in surveillance and other real world applications. You will also learn that while Frontal-Frontal accuracy of face recognition algorithms is above 95%, the currently best Frontal-Profile algorithms achieve 85% accuracy. That is merely a 10% loss, which given the problem, is quite remarkable. So yes, one can recognize a person with decent accuracy from the profile without knowing the profile in advance.


Wow. Did you think nobody would read the article? You're seriously comparing a bunch of red lines you drew to the sophisticated techniques outlined here, and then claiming your success rate--85% in the best case--should be the same?

Do your red lines do any of the following:

HoG
: We extract square patches of width 10, 15, 30, 50 pixels centered around each of the 30 facial key-points. Then we extract HoG features of cell-size 8 from these patches and concatenate them to form a 53k dimensional HoG feature of the face. Multiple-scale
patches are used to provide a multi-resolution view of the face. We use the VLFeat [34] implementation of HoG.

LBP
: Similar to HoG we extract square patches of size 10, 15, 30, 50 and 100 pixels centered around 30 key-points. We then extracted uniform LBP features (sampling points 16) of radius 1 and concatenate them to form a 36k dimensional LBP feature of the face.

Fisher Vector
: We used publicly available code of Fisher Vector and followed the same principle of [31]. However we didn’t use horizontal flipping of images to make it consistent with other features. Fisher vector encoding with 512 cluster centers result in a 67,584 dimensional feature.

Deep features
: We use the trained network reported in [9]. The authors use a deep network with 10 convolution layers, 5 pooling layers and 1 fully connected layer. The receptive field of the CNN is 100×100×1
.
The authors claim that a deeper network with a smaller number of filters is easier to train because it uses fewer parameters and performs better due to high amount of non-linearity. The network is trained on the CASIA-Webface data set [37] with 494,414 images of 10,575 subjects. We only used the network to extract features of dimension 320. We used a simple Cosine similarity measure over this feature.


Roger M.
Posts: 1598
Joined: January 17th, 2008, 12:00 pm

Re: ERDNASE

Postby Roger M. » May 5th, 2018, 10:43 am

But Chris hand drew some random red lines with a paint program that came pre-installed on his computer, isn't that the same thing?

User avatar
jkeyes1000
Posts: 483
Joined: August 2nd, 2012, 3:12 pm

Re: ERDNASE

Postby jkeyes1000 » May 5th, 2018, 11:00 am

Jack Shalom wrote:
lybrary wrote:
Since Bill's pseudo-science knowledge is so dishonest I suggest for anybody who is genuinely interested in this subject to read for example this article http://www.rci.rutgers.edu/~vmp93/Confe ... 16_CFP.pdf
What you will learn is that recognizing profile images from frontal images is a common problem in surveillance and other real world applications. You will also learn that while Frontal-Frontal accuracy of face recognition algorithms is above 95%, the currently best Frontal-Profile algorithms achieve 85% accuracy. That is merely a 10% loss, which given the problem, is quite remarkable. So yes, one can recognize a person with decent accuracy from the profile without knowing the profile in advance.


Wow. Did you think nobody would read the article? You're seriously comparing a bunch of red lines you drew to the sophisticated techniques outlined here, and then claiming your success rate--85% in the best case--should be the same?

Do your red lines do any of the following:

HoG
: We extract square patches of width 10, 15, 30, 50 pixels centered around each of the 30 facial key-points. Then we extract HoG features of cell-size 8 from these patches and concatenate them to form a 53k dimensional HoG feature of the face. Multiple-scale
patches are used to provide a multi-resolution view of the face. We use the VLFeat [34] implementation of HoG.

LBP
: Similar to HoG we extract square patches of size 10, 15, 30, 50 and 100 pixels centered around 30 key-points. We then extracted uniform LBP features (sampling points 16) of radius 1 and concatenate them to form a 36k dimensional LBP feature of the face.

Fisher Vector
: We used publicly available code of Fisher Vector and followed the same principle of [31]. However we didn’t use horizontal flipping of images to make it consistent with other features. Fisher vector encoding with 512 cluster centers result in a 67,584 dimensional feature.

Deep features
: We use the trained network reported in [9]. The authors use a deep network with 10 convolution layers, 5 pooling layers and 1 fully connected layer. The receptive field of the CNN is 100×100×1
.
The authors claim that a deeper network with a smaller number of filters is easier to train because it uses fewer parameters and performs better due to high amount of non-linearity. The network is trained on the CASIA-Webface data set [37] with 494,414 images of 10,575 subjects. We only used the network to extract features of dimension 320. We used a simple Cosine similarity measure over this feature.



I assume that the pictures used in this study were of the modern (digital) kind. Such resolution is impossible to attain in poorly reproduced lithographs or offset prints of photographs taken in 1912 (or whenever it was).

You can't blame Chris for that now--can you?

Such scrutiny as the cited article recommends would be a waste of time in this case.

User avatar
lybrary
Posts: 1169
Joined: March 31st, 2013, 4:59 pm
Contact:

Re: ERDNASE

Postby lybrary » May 5th, 2018, 11:21 am

Please guys, what do most of you understand about neural networks? I have programmed, developed and used neural networks when it wasn't called deep learning and fashionable to use them. My illustration with the red lines is there so that it can be intuitively understood by everybody at least those with common sense. You guys have none. When I write about stylometry I also don't dive into the mathematical details of Ward's method, cosine metrics, complete or partial linking, hierarchical clustering, support vector machines, or perceptrons. It is pointless to talk about the mathematical foundation of these methods here. I have that discussion with the researchers involved in developing these method.
Lybrary.com Magic & Gambling
preserving magic one book at a time

Jackpot
Posts: 236
Joined: June 8th, 2016, 12:38 am
Location: Santa Rosa, CA

Re: ERDNASE

Postby Jackpot » May 5th, 2018, 11:56 am

jkeyes1000 wrote:I assume that the pictures used in this study were of the modern (digital) kind. Such resolution is impossible to attain in poorly reproduced lithographs or offset prints of photographs taken in 1912 (or whenever it was).

You can't blame Chris for that now--can you?

Such scrutiny as the cited article recommends would be a waste of time in this case.


Dr. Wasshuber presents the cited article as "evidence" for his argument. Unfortunately the article he has chosen does not support the methods he used due to the technological limitations you reference. If he had selected a more appropriate article, one which relied on the facial recognition techniques he used and the quality of the photographs available, we might have some idea as to the true accuracy of his identification of Gallaway from the photographs.

Since the scrutiny the article recommends is impossible with the photographs available we are faced with the same problem that exists with Olsson's report. Dr. Wasshuber presented the literary analysis as definitive, but it is not. It could not rule out any of the candidates due to the limitations of the analytical process he relied on. In this latest case the photo analysis does not rule out a large number of the male population of the United States at the time the picture was taken.
Not the one who created the Potter Index.

Roger M.
Posts: 1598
Joined: January 17th, 2008, 12:00 pm

Re: ERDNASE

Postby Roger M. » May 5th, 2018, 12:07 pm

lybrary wrote:I have programmed, developed and used neural networks when it wasn't called deep learning and fashionable to use them. My illustration with the red lines is there so that it can be intuitively understood by everybody at least those with common sense. You guys have none. When I write about stylometry I also don't dive into the mathematical details of Ward's method, cosine metrics, complete or partial linking, hierarchical clustering, support vector machines, or perceptrons. It is pointless to talk about the mathematical foundation of these methods here. I have that discussion with the researchers involved in developing these method.


In this post, Chris is telling you you're all far too stupid to understand what he's saying.

The "I'm far smarter than you, therefore you can't possibly understand what I'm saying" ploy is a great way to mask utter bulls_it, which is exactly what Chris is doing in this post.

Bob Coyne
Posts: 717
Joined: January 26th, 2008, 12:00 pm
Favorite Magician: Charlies [sic]
Location: New York, NY

Re: ERDNASE

Postby Bob Coyne » May 5th, 2018, 12:18 pm

lybrary wrote:Please guys, what do most of you understand about neural networks? I have programmed, developed and used neural networks when it wasn't called deep learning and fashionable to use them. My illustration with the red lines is there so that it can be intuitively understood by everybody at least those with common sense. You guys have none. When I write about stylometry I also don't dive into the mathematical details of Ward's method, cosine metrics, complete or partial linking, hierarchical clustering, support vector machines, or perceptrons. It is pointless to talk about the mathematical foundation of these methods here. I have that discussion with the researchers involved in developing these method.

I'm confused -- so what have you actually done? You're saying that the red lines argument isn't your actual method and that you've trained a neural network that you're applying to this instance? If so, how was it trained? And what sort of accuracy does it have on similar examples?

User avatar
jkeyes1000
Posts: 483
Joined: August 2nd, 2012, 3:12 pm

Re: ERDNASE

Postby jkeyes1000 » May 5th, 2018, 12:45 pm

I see no fault in Chris's method of Facial Recognition (although I do concur with Bill when he suggests that the hair on the head ought not to be aligned with the bald pate of another subject).

The authors of the article that Chris refers to would be amongst the first to say that their techniques must be modified according to the quality of the images being examined.

I think Chris is right to advise "common sense". It is simply a matter of doing the best we can with what evidence we have.

User avatar
lybrary
Posts: 1169
Joined: March 31st, 2013, 4:59 pm
Contact:

Re: ERDNASE

Postby lybrary » May 5th, 2018, 12:47 pm

Jackpot wrote:In this latest case the photo analysis does not rule out a large number of the male population of the United States at the time the picture was taken.
Yes, it does rule out a large number of the male population in the US at that time. The starting pool isn't all males. The starting pool are operators of the Monotype keyboard in the US. That is a pool on the order of ~1000 people in 1912, likely less. It was an emerging technology.

Roger M. wrote:In this post, Chris is telling you you're all far too stupid to understand what he's saying.
If you can't look at the photos I have presented and follow the red lines, then yes, I am afraid I can't further help you in your grasp of this subject. If you do want to understand the algorithmic and mathematical details I have provided a starting point. If the environment here wouldn't be as toxic as it is more would be possible.
Lybrary.com Magic & Gambling
preserving magic one book at a time

Brad Henderson
Posts: 4546
Joined: January 17th, 2008, 12:00 pm
Location: austin, tx

Re: ERDNASE

Postby Brad Henderson » May 5th, 2018, 1:01 pm

lybrary wrote:Please guys, what do most of you understand about neural networks? I have programmed, developed and used neural networks when it wasn't called deep learning and fashionable to use them. My illustration with the red lines is there so that it can be intuitively understood by everybody at least those with common sense. You guys have none. When I write about stylometry I also don't dive into the mathematical details of Ward's method, cosine metrics, complete or partial linking, hierarchical clustering, support vector machines, or perceptrons. It is pointless to talk about the mathematical foundation of these methods here. I have that discussion with the researchers involved in developing these method.



bull [censored]

first. loose the condescension.

second, bull [censored].

i

Brad Henderson
Posts: 4546
Joined: January 17th, 2008, 12:00 pm
Location: austin, tx

Re: ERDNASE

Postby Brad Henderson » May 5th, 2018, 1:03 pm

jkeyes1000 wrote:The authors of the article that Chris refers to would be amongst the first to say that their techniques must be modified according to the quality of the images being examined.


ah. we are playing the mind reading game are we?

User avatar
Richard Kaufman
Posts: 27047
Joined: July 18th, 2001, 12:00 pm
Favorite Magician: Theodore DeLand
Location: Washington DC
Contact:

Re: ERDNASE

Postby Richard Kaufman » May 5th, 2018, 1:46 pm

Some temporary bans will hopefully allow some people to catch their breath.
Subscribe today to Genii Magazine

Roger M.
Posts: 1598
Joined: January 17th, 2008, 12:00 pm

Re: ERDNASE

Postby Roger M. » May 6th, 2018, 11:07 am

Bob Coyne wrote:I'm confused -- so what have you actually done? You're saying that the red lines argument isn't your actual method and that you've trained a neural network that you're applying to this instance? If so, how was it trained? And what sort of accuracy does it have on similar examples?


Or is Chris is saying that the fundamental principals of neural networks and deep learning are the underlying principals in his entire search for Erdnase?

Jackpot
Posts: 236
Joined: June 8th, 2016, 12:38 am
Location: Santa Rosa, CA

Re: ERDNASE

Postby Jackpot » May 6th, 2018, 12:00 pm

jkeyes1000 wrote:The authors of the article that Chris refers to would be amongst the first to say that their techniques must be modified according to the quality of the images being examined.

I think Chris is right to advise "common sense". It is simply a matter of doing the best we can with what evidence we have.


Agreed. But significantly more important than any agreements that techniques must be modified would be the authors' informed opinions as to the validity of the the conclusions reached by applying a crude facial recognition method to photographs of a lesser quality than those they use.

I wonder if the investigator applied the red line and ruler facial recognition method (rl&rfrm) to the entire "pool on the order of ~1000 people in 1912", or if he only compared the rl&rfrm photographs to the single picture of Gallaway. For results to be valid common sense dictates that the comparisons be made to the entire pool of ~1000 people.
Not the one who created the Potter Index.

Bob Coyne
Posts: 717
Joined: January 26th, 2008, 12:00 pm
Favorite Magician: Charlies [sic]
Location: New York, NY

Re: ERDNASE

Postby Bob Coyne » May 7th, 2018, 8:17 am

I've added a couple dozen interesting new linguistic correspondences to my Erdnase-Sanders document. These are marked [New...] or [Updated...]. The highlights section has also been augmented to hold more of the burgeoning set of "best" comparisons.

http://www.cs.columbia.edu/~coyne/erdnase-sanders-use-of-language.html

As part of this, I recently examined a couple new sources of Sanders' writing -- his Masters Thesis and a summer mining expedition report he made while in college. While dry and technical (like his other mining writings), I found various new extracts that closely parallel Erdnase, some to reinforce existing examples and some supplying new ones. I marked these sources when quoted as [THESIS] and [SMR].

In addition, I discovered some very nice new correspondence in the previous sources I've looked at. Here's one I think is highly significant (in particular the common analogy made at the end that I don't think I've seen mentioned before):

---- Sanders on GAMBLING (poem and bio to Huntington) ---------
In this poem and bio, mining and card playing are tightly linked (as they were historically). Huntington (one of Sanders' classmates) left mining to settle down with a family and work in education. Sanders refers to it as having "quit the game." Erdnase also references that same phrase. And then Sanders says that Huntington is "STRAIGHT and true."

Sanders:
So, Huntington, you QUIT THE GAME
Our mining engineers HAVE PLAYED
Erdnase: In most card GAMES ... there is an old adage much quoted that runs, "If suspected, QUIT."

-----straight (honest)
Sanders: how thoughtful, gentle, STRAIGHT and true

And here's the kicker: both Sanders and Erdnase explicitly contrast other MORE RESPECTABLE PROFESSIONS (education and stock trading) with the tabooed WILD DELIGHTS and SENSATIONS associated with gambling and/or mining. The two men make the exact same characterization in much the same manner!

Sanders: Huntington has placed taboo the WILD DELIGHTS AND EXHILERATING INFLUENCES of the MINING PROFESSION and settled into the more prosaic, even if MORE RESPECTABLE, calling of education.
Erdnase: have impressed the PROFESSIONAL CARD PLAYER with a certain knowledge that his MORE RESPECTED brother of the stock exchange possesses
Erdnase: Hazard at play carries SENSATIONS THAT ONCE ENJOYED ARE RARELY FORGOTTEN


Return to “General”