|This article must adhere to the biographies of living persons policy, even if it is not a biography, because it contains material about living persons. Contentious material about living persons that is unsourced or poorly sourced must be removed immediately from the article and its talk page, especially if potentially libellous. If such material is repeatedly inserted, or if you have other concerns, please report the issue to this noticeboard. If you are connected to one of the subjects of this article and need help, please see this page.|
|This article is of interest to the following WikiProjects:|
- 1 Recent Honors / Interviews / Videos
- 2 A self-promoting article
- 3 General
- 4 Godel/Goedel/Go-doubledot-del Machine
- 5 Updates / Corrections
- 6 Vanity or the opposite?
- 7 Notability
- 8 Merge
- 9 Low-Complexity Art and Theory of Beauty & Interestingness & Curiosity
- 10 Notability tag
Recent Honors / Interviews / Videos
I know that biographies of living persons are a delicate issue, but this one seems to be a bit out of date. For example, since 2009 he is Professor of AI in Switzerland. And the text does not mention that his group now has the best systems for connected handwriting recognition, based on recurrent neural networks, and also for traditional handwriting recognition, based on deep neural networks, see http://arxiv.org/abs/1003.0358
The article should also mention recent honors:
Elected to the European Academy of Sciences and Arts (2009)
Best paper awards: AGI 2010 best paper award, GECCO 2009 best paper award, GECCO 2005 best paper award
One could insert links to more recent interviews and articles:
1. Build An Optimal Scientist, Then Retire (2010): http://www.hplusmagazine.com/articles/ai/build-optimal-scientist-then-retire
2. Slashdot on Schmidhuber's artificial curiosity (2010): http://developers.slashdot.org/story/10/01/28/0052202/Can-Curiosity-Be-Programmed
3. Gödel’s Gift (2010): http://www.thefifthconference.com/topic/tech/goedel%E2%80%99s-gift
Some of the Wikipedia biographies have links to video lectures; one could add his recent talk at the Singularity Summit 2009 at Vimeo http://www.vimeo.com/7441291 or youtube: http://www.youtube.com/watch?v=Ipomu0MLFaI
A self-promoting article
I think this is just a self-promoting article of people that like Schmidhuber ideas (former students of him, or even himself, since much of it is just propaganda and copy/paste from his own webpage). The length of the article is just disproportionate compared to who is him and his relevance in the academic world. he should be barely mentioned. Some of his ideas are popular science, yet they seem to me endorsed by this Wikipedia article. A pity. I propose to do something about it. 188.8.131.52 (talk) 18:36, 26 November 2009 (UTC)
Contents overview: Stan's Brain Power Query has a fun comment on Alzheimer and Schmidhuber's comparison of human brains and computers. Kosebamse is not satisfied with a cryptic four sentence summary of Schmidhuber's recent preprint on the Godel machine. The rest of the discussion so far is mostly a rather redundant exchange between Paul and Newbie. Paul first claimed he found an essential flaw in the Godel machine, mentioning real numbers and Cantor and other things without going into details. Newbie thinks he fully understands the Godel machine, and says that it is theoretically sound and that the non-computability of certain real numbers is not a problem in this context, and repeatedly challenged Paul to be more specific. Paul is annoyed that Newbie refuses to reveal his true indentity, but toned down his claims a bit, although he says he is still trying to work on a counterexample for Theorem 2.1. The entire page is a bit of a mess; several postings do not appear in chronological order, so one has to check the history information to figure out who said what when. Newbie 21:29, 12 Mar 2004 (UTC)
Who am I?
Am I in a phone directory, white pages or a whoIsWho?
Say who you are and I will tell you. Paul Beardsell 07:10, 20 Feb 2004 (UTC)
Brain Power Query
Mr Schmidhuber contends that by the time he retires he will be able to afford to buy a computer that contains more computing power than his brain, though he doesn't indicate whether he expects to have contracted Alzheimer's by that time. Is there a measure of computing power that equates to a measure of brain power? He makes a stab at it at http://www.idsia.ch/~juergen/raw.html, but I'm not sure he's right. I don't have a floating point processor in my brain and still use my fingers for counting. What ought to be possible to calculate is whether the amount of power consumed by the brain equates to the amount of power used by a computer. The measure is watts. I'm particularly interested in whether the human brain has a hibernate mode (like some laptops), i.e. in whether the brain's sugar consumption - which I gather is considerable in relation to the rest of the human body - depends on the quantity, quality or nature of thought. Just think, we could market the Wikipedia diet! Any thoughts (Paul especially)? Stan 19:40, 5 Mar 2004 (UTC) <-- To introduce myself, I am an erstwhile colleague of Paul Beardsell. (And apologies if my irreverent posting has nothing to to with the rest of the discussion on this page.)
Well, Matthew, I have lost a little weight recently. Usable power isn't only Watts consumed. It depends on efficiency, essentially the difference in temperature, also. Presumably that's why keeping a cool head is considered good, and being hot-headed bad.
I find myself a little bit in the position of the little boy in the Emperor without any Clothes story. At first I did not rudely shout out (like the little boy) but whispered by e-mail to the emperor that I thought his fly was down. But I am pretty confident I can construct the necessary counterexample: I'm just working it through again. I am trying to do it without resorting to the psuedo-Maths in the paper. I mean, <curly-italic-capital-E> isn't even defined. Not that this bothers Newbie. Maybe it is a widely understood term. But Cantor is mentioned needlessly, so why bother defining terms? Paul Beardsell 04:15, 10 Mar 2004 (UTC)
Wot got deleted
Moved from the article:
The machine essentially is a traditional computer whose initial software includes a systematic generator of proofs and programs. It also includes axioms describing hardware and initial software and some formal performance measure. Some generated software-rewriting program is executed provided the theorem prover also generated a proof that this rewrite will improve the machine's performance. It is easy to see that such rewrites cannot correspond to local minima - they must be globally optimal.
This looks a little incomprehensible. Could somebody please explain this for the laypeople? Kosebamse 15:01, 2 Mar 2004 (UTC)
Schmidhuber's home page
This whole article is a copy from Juergen Schmidhuber's home page. Without mention of any express permission. That is without question. Questionable is my opinion that Goedelman is Juergen Schmidhuber. I have good reason to believe that the Goedel Machine referenced is without good foundation. Certainly the paper is not hard science as I understand it. It qualifies as speculation only. Interestingly someone else has recently removed the Goedel Machine article for copyright violation. Which is a pity as I was going to post correspondence between Schmidhuber and me on its Talk page. Maybe I should do so here. Paul Beardsell 15:12, 2 Mar 2004 (UTC)
In Theory, at least
I am a theoretical computer scientist. I read the Godel machine paper several times and could not find any serious flaw. I think it is rather exciting. The moved quote above is too compact though. Newbie 20:05, 3 Mar 2004 (UTC)
Please expand the quote so that it is understandable. Paul Beardsell 04:24, 10 Mar 2004 (UTC)
Non-Turing machine necessary?
Please explain the paragraph in italics above. Its all gobbledygook to me (and I know some computer science). I suggest you start with the sentence It is easy to see... Then I intend to quiz you about the assertion in the paper that a non-Turing machine will be necessary to get the Goedel Machine up and running but that it can all be simulated on a PC. You did read that bit? Paul Beardsell 22:47, 3 Mar 2004 (UTC)
For the record, I too find the idea of the Goedel Machine exciting. It was the profound disappointment I experienced upon reading the paper that has got me riled up. And, as I am a suspicious kind of guy, were Newbie to care to identify himself I would be more inclined to give his views the weight they might deserve but which, as this is his one and only posting to Wikipedia so far... Paul Beardsell 08:43, 4 Mar 2004 (UTC)
No non-Turing machine necessary
My weight (if any) does not matter; the paper's content does. It does not say a non-Turing machine is necessary. That would be totally against its spirit. On page 5 it even has a paragraph on a Turing machine as an example hardware. It just says you don't need a universal Turing machine. You may instead use a more limited finite state automaton such as a PC. The important thing is that its software contains axioms describing the hardware, including storage limitations. I guess the easy to see refers to the 1 paragraph proof of Theorem 2.1. Newbie 20:21, 4 Mar 2004 (UTC)
I deny that Theorem 2.1 is proved in Schmidhuber's paper. I think a counterexample is trivial. Paul Beardsell 08:33, 8 Mar 2004 (UTC)
Then why don't you give it? Newbie 19:40, 8 Mar 2004 (UTC)
I could do so and do not. Why not? You could identify yourself but do not. Why not? Paul Beardsell 03:01, 9 Mar 2004 (UTC)
That's like justifying your failure to deliver the counterexample by saying "You could dance on the table but do not. Why not?" Because it's not relevant to the present discussion. The only thing that counts are the facts. Where are your facts? Newbie 19:57, 9 Mar 2004 (UTC)
It is similar but not the same. Had I criticised a dance, you had defended it saying you were a qualified dancer, but refused to demonstrate any dance, then I could ask your name to find out if you are indeed a dancer. Here you say you are a theoretical computer scientist, you assert that a computer science paper is good science (having read it carefully several times), you refuse to enter into computer science argument, you say I should back up my claims. Well, who to? To someone who won't understand my argument? To someone who, if he is shown to be wrong, has lost nothing? No. Here I am, readily identifiable putting my possibly inconsiderable reputation on the line. Who are you? And how long have you worked at IDSEA? Paul Beardsell 01:23, 10 Mar 2004 (UTC)
Your response looks like a rather weak attempt to wriggle out of the mess you created for yourself - I am sure you see that there is a big difference here. You use this public site to claim Schmidhuber is wrong. Your potential audience is everybody, not just me. I don't believe your claim, and challenge you to back it up by facts. You ask, who to? Answer: to anybody who's reading this, of course, not just me. It does not matter whether I am a theoretical computer scientist (ignore this irrelevant claim), or whether I have worked at IDSEA or MIT or McDonald's drive thru. The only thing that counts are the facts; scientific claims are not settled by authority but by facts and logic. You write I refuse to enter a computer science argument, when in fact for the umpteenth time I ask you to make a precise statement or formal claim that we can discuss at all. As you say, you did put your reputation on the line. Now you have to deliver, or minimize loss of face by admitting you were wrong. Newbie 20:04, 10 Mar 2004 (UTC)
But when I persuasively demonstrate a central flaw in the paper, by now I am quite keen that you should lose some face. You have stated that you have read the paper several times, you seem to have no problem understanding it, yet you decline to explain even the paragraph quoted here which you say is perfectly understandable but a little too dense. You assert the paper is good science. Indeed, you could find nothing wrong with it. It isn't only me who is exposed here. It is me, identified, and you, anonymous. By now you have started to make attacks on me, perhaps provoked. But you can just slink away, afterwards, when the Godel Machine is shown to be improperly argued in Schmidhuber's paper.
I'm working on my task. You explain the paragraph and say who you are. If you object to one you can still do the other, or so you say.
Paul Beardsell 00:02, 12 Mar 2004 (UTC)
Well, you maneuvered yourself into this awkward and (as you point out) asymmetric position. Now you must either retreat or deliver your publicly announced counterexample, no matter whether I slink away or not. But rest assured: I'll be the first to admit I was wrong in case you can formally demonstrate an essential flaw in Schmidhuber's reasoning. Although in that case you won't need my applause to improve your general standing; I am sure you'll get the attention of others as well. Given the nature of your previous statements, however, I am afraid this case is quite unlikely. Nevertheless, after all this rather fruitless exchange so far, I am now actually playing with the thought of following your request for more explanation, by writing a full Godel machine entry for Wikipedia. But don't wait for it, continue working on your counterexample! Newbie 20:43, 12 Mar 2004 (UTC)
Newbie, as you claim to be a theoretical computer scientist, you will know that a Turing Machine cannot cope with the set of Real numbers. Schmidhuber's paper and his "easy to see" theorem makes use of the set of Real numbers. What say you? Paul Beardsell 07:37, 8 Mar 2004 (UTC)
I ask: Is that your counterexample announced above? First you claim the paper says a non-Turing machine is necessary. Obviously not true. Then you make a vague statement about Turing machines and real numbers (possibly referring to the fact that most real numbers are incomputable?). You seem to suggest this somehow exhibits a flaw in Theorem 2.1 (if you don't want to download the entire pdf see http://www.idsia.ch/~juergen/gmweb3/node10.html ). But you don't say why. What do you mean? Try to be precise. Newbie 19:40, 8 Mar 2004 (UTC)
No, that is not my counterexample. Which is one of the reasons it is in a separate section. Here we have a paper that goes to the trouble of telling us it knows about Cantor's "trick" used by Godel but never mentions Cantor again, that goes to the trouble of letting us know some details of the Turing machine which are not required elsewhere in the paper, that goes to the trouble of doing many unnecessary things, as nothing but padding. Yet it uses the set of Real numbers without a care in the world, without a word of explanation as to why this might not matter. And it might not. But why not? Also pertinent: I reckon he need not have used the Real numbers at all! Why are you, as a theoretical computer scientist, not making that point? Paul Beardsell
Ok, where is your counterexample? You say it's in a separate section, but there is none. The rest of your reply also is vague and does not contribute anything to substantiating your claims. What exactly is your problem with the real numbers??? Newbie 19:57, 9 Mar 2004 (UTC)
As I said, and you confirmed, not-computable. Therefore it is not helpful to use them in a proof which says something about Turing machines. At the least an explanation is required. So strike out the Cantor name-dropping, if there is a shortage of space, and do so. Paul Beardsell 01:15, 10 Mar 2004 (UTC)
Re-reading the above, I am now convinced that you don't fully understand the concept of a proof. A proof is a finite sequence of symbols. But it may talk about incomputable objects. For example, the theorem forall x in R: x=x+1-1 says something about incountably many real numbers, most of them incomputable. Nevertheless, this theorem can be derived from a few axioms within a finite number of steps. In particular, the proof is computable on a Turing machine (should be obvious to anybody with a computer science education). Since you never clearly formulated your criticism, I have to guess what kind of explanation you might consider useful, but in the light of your vague claims above, maybe this is it. Newbie 20:04, 10 Mar 2004 (UTC)
What a happy result! As a by-product of the Godel Machine, Schmidhuber may have solved the Artificial Consciousness problem:
- Its bootstrap mechanism is based on a simple idea. Its initial software or program p includes an axiomatic description of (possibly stochastic) environmental properties and of the Godel machine’s goals and means. 2nd para 1.2 Basic Principles.
This is just the starting point of the Godel machine!
Paul Beardsell 07:58, 8 Mar 2004 (UTC)
Gödel Machine at Scholarpedia
I read the rather bizarre old discussion above (made smaller headings), and found a text about the Gödel Machine at Scholarpedia: http://www.scholarpedia.org/article/Universal_search Why not simply use this? I cut and paste:
- "The Gödel Machine (Schmidhuber, 2006) is a general paradigm for solving arbitrary problems, including optimization problems and reinforcement learning. Inspired by the work of Kurt Gödel, it is based on a set of axioms and a programming language for encoding and deriving formal proofs. The machine interacts with an environment, through a dedicated hardware and software system, and its aim is to maximize a reward over a possibly limited lifetime. The axioms include a detailed formal description of the machine's software and hardware, including both the components interacting with the environment and those dealing with the formal proofs, and a possibly partial description of the environment. The Gödel machine starts its interaction with the environment according to some initial program; in the meantime, a proof search algorithm, is used to find provably optimal modifications of the machine's software. In other words, a Gödel machine can rewrite any part of its software, but only after it can prove that the modification will increase the expected reward for the remaining lifetime: in this case, meta-learning can influence any aspect of the machine's behavior, including the proof search itself, by rewriting the code controlling it. (Schmidhuber, 2006) also describes example instantiations of a Gödel Machine, in which the initial proof search algorithms are variants of universal search and OOPS, respectively."
The Scholarpedia article also has additional references. The biography is rather long - instead of pasting this here, better use it to start a new Wikipedia article? Fleabox (talk) 19:36, 22 August 2008 (UTC)
Updates / Corrections
One should mention that he is also professor of cognitive robotics at TU Munich. Should have been inserted more than a year ago!
184.108.40.206 (apparently some SBC address in the US) edited the paragraph ending: This work on digital physics also led to limit-computable generalizations of algorithmic information and the concept of Super Omegas, which are limit-computable numbers that are even more random (in a certain sense) than Gregory Chaitin´s number of wisdom Omega.
To this 220.127.116.11 added: However, his theories are not without controversies, as no good computer implementations exists that would solidify his claims. Deep issues in machine learning and statistics, such as the bias variance dilemma, find insufficient treatment in Schmidhuber's writings.
This seems to be complete nonsense. Omega is an incomputable but enumerable number, and Schmidhuber's Super-Omega is not even enumerable, although it is limit-computable. All his Omega results are totally theoretical, and were published in a journal of theoretical computer science. Obviously this statement: no good computer implementations exists that would solidify his claims just indicates that 18.104.22.168 has no idea what he/she is talking about. The bias-variance dilemma also has nothing to do with this. Clearly, this should be removed. IDSIAupdate 09:07, 22 February 2006 (UTC)
Vanity or the opposite?
Shouldn't this entry be deleted according to Wikipedia's vanity guidelines? Most of the claims are highly dubios at best, e.g. "One ambitious theoretical contribution is his 30-page preprint (2003) on the as yet unrealized Gödel machine which, we are told, would solve arbitrary computational problems in an optimal fashion inspired by Kurt Gödel's celebrated self-referential formulas (1931)." (Anonymous posting by 22.214.171.124)
Isn't that the opposite of vanity? Sounds rather negative, especially the part "we are told"... Someone apparently inserted this years ago when the only Goedel machine publication was a tech report (see strange discussion above). But yes, this sentence should be removed at least as long there is no decent article about the Goedel machine. Will do it. Instead one should mention the important general topic of universal learning algorithms, and also create an entry for his coworker Hutter. IDSIAupdate 08:59, 17 March 2006 (UTC)
Well, there are many independent articles that mention his work and build upon it. For example, if you go to Google Scholar and type in "J Schmidhuber" you can follow the links to many citations by others: http://scholar.google.com/scholar?q=j+schmidhuber&hl=en&lr= . And this search actually fails to find some of the most cited articles. For example, the "Long Short-Term Memory" paper (Hochreiter & Schmidhuber, 1997) does not appear among the results, although Google Scholar does know it: http://scholar.google.com/scholar?hl=en&lr=&q=%22long+short+term+memory%22 . (Some bug in the search algorithm?) Anyway, one could add a few links of this type. IDSIAupdate 09:08, 28 February 2007 (UTC)
About the question itself:
- I'd first say this page is excessively big compared to the notability of the subject. I went to look at pages about, say, Robert Tarjan (who was awarded the Turing award), and the articles are of comparable length. I don't think that J. Schmidhuber is of comparable importance, and I know some professors in University of Catania with production of comparable impact.
- Indeed, he has published peer-reviewed work, with an h-index of around 18 (manually computed from Google Scholar results, but they); however, in this context (see WP:ACADEMIC) "independent sources" means "independent biographies", i.e. an reliable indipendent source stating his importance. We are simply not allowed to decide ourself that this scientist is notable, without such sources. From WP:BRAIN (which is a synthesis of WP:SYN):
- "Your brain can put two facts together to create new facts, but these new facts do not belong on Wikipedia "Main" namespace. They must be sourced.".
I just tried your link (Wed Aug 13) but got a Google h-index of 30, not 18. That is, 30 papers with at least 30 citations, one of them with 274. You could insert this link http://scholar.google.com/scholar?q=j+schmidhuber&hl=en&lr= and get thousands of external sources. That by itself would certainly satisfy WP:ACADEMIC. Alternative criteria are satisfied as well: the collective body of his work is significant and well-known, and he is regarded as an important figure by independent notable academics, otherwise they wouldn't invite him to give all these keynote talks (from the CV: ICANN 2008, KES 2008, Cog. Systems 2008, ALT 2007 & DS 2007 joint invited lecture, A*STAR 2007, ACAT 2007, Art Meets Science 2007, Zuse Symposium 2006, GWAL 2006, Turing Days 2006, ICANN 2005...). Personally I do not believe in h-indexes; to me the essential thing is that he introduced important new concepts such as his theory of aesthetics and beauty and interestingness and artificial curiosity, which I am familiar with (recently on TV). Fleabox (talk) 20:26, 13 August 2008 (UTC)
I found a German 3 page article on Schmidhuber in CIO magazine: "Der ideale Wissenschaftler" meaning: the ideal scientist. http://www.cio.de/karriere/personalfuehrung/803246/ and another one (2 pages) on simulated universes http://www.idsia.ch/~juergen/AargauerUniverse.pdf One could add such stuff to the biography, together with this Scholarpedia article: http://www.scholarpedia.org/article/Universal_search Fleabox (talk) 19:56, 22 August 2008 (UTC)
Low-Complexity Art and Theory of Beauty & Interestingness & Curiosity
I tried to improve the low-complexity art article. The biography mentions low-complexity art and related concepts, but I find it wanting. I am not sure though what should go in the biography and what should go in the specialized articles. To summarize, his algorithmic theory of beauty takes the subjectivity of the observer into account: among several observations classified as comparable by a given subjective observer, the most beautiful one is the one with the shortest description, given the observer’s previous knowledge and his particular method for encoding the data. This is closely related to the principles of algorithmic information theory and minimum description length. One of his examples: mathematicians enjoy simple proofs with a short description in their formal language. Another example describes a pretty human face whose proportions can be described by very few bits of information, drawing inspiration from less detailed 15th century proportion studies by Leonardo da Vinci and Albrecht Dürer. But Schmidhuber's theory explicitly distinguishes between what's beautiful and what's interesting, stating that interestingness corresponds to the first derivative of subjectively perceived beauty, assuming that the observer continually tries to improve the predictability and compressibility of his observations by discovering regularities such as repetitions and symmetries and fractal self-similarity. Whenever the observer's learning process (such as a predictive neural network) leads to improved data compression such that the observation sequence can be described by fewer bits than before, the temporary interestingness of the data corresponds to the number of saved bits. This compression progress is proportional to the observer's internal reward, also called curiosity reward. A reinforcement learning algorithm can be used to maximize future expected reward by learning to execute action sequences that cause additional interesting input data with yet unknown but learnable predictability or regularity. The principles can be implemented on artificial agents which then exhibit a form of artificial curiosity.
- J. Schmidhuber. Low-complexity art. Leonardo, Journal of the International Society for the Arts, Sciences, and Technology, 30(2):97–103, 1997. http://www.jstor.org/pss/1576418
- J. Schmidhuber. Papers on the theory of beauty and low-complexity art since 1994: http://www.idsia.ch/~juergen/beauty.html
- J. Schmidhuber. Facial beauty and fractal geometry. Cogprint Archive: http://cogprints.soton.ac.uk , 1998
- J. Schmidhuber. Simple Algorithmic Principles of Discovery, Subjective Beauty, Selective Attention, Curiosity & Creativity. Proc. 10th Intl. Conf. on Discovery Science (DS 2007) p. 26-38, LNAI 4755, Springer, 2007. Also in Proc. 18th Intl. Conf. on Algorithmic Learning Theory (ALT 2007) p. 32, LNAI 4754, Springer, 2007. Joint invited lecture for DS 2007 and ALT 2007, Sendai, Japan, 2007. http://arxiv.org/abs/0709.0674
- J. Schmidhuber. Curious model-building control systems. International Joint Conference on Neural Networks, Singapore, vol 2, 1458–1463. IEEE press, 1991
- J. Schmidhuber. Papers on artificial curiosity since 1990: http://www.idsia.ch/~juergen/interest.html
- J. Schmidhuber. Developmental robotics, optimal artificial curiosity, creativity, music, and the fine arts. Connection Science, 18(2):173–187, 2006
- Schmidhuber's theory of beauty and curiosity in a German TV show: http://www.br-online.de/bayerisches-fernsehen/faszination-wissen/schoenheit--aesthetik-wahrnehmung-ID1212005092828.xml