tag:blogger.com,1999:blog-319304714858376628.comments2020-08-29T10:27:37.070+02:00The Robust Mathematical Modeling BlogIgorhttp://www.blogger.com/profile/17474880327699002140noreply@blogger.comBlogger9125tag:blogger.com,1999:blog-319304714858376628.post-72159507316994314152012-02-16T15:44:47.152+01:002012-02-16T15:44:47.152+01:00of related interest:
Odds Are, It's Wrong
ht...of related interest: <br /><br />Odds Are, It's Wrong<br />http://www.sciencenews.org/view/feature/id/335872/title/Odds_Are%252C_Its_WrongIgorhttps://www.blogger.com/profile/17474880327699002140noreply@blogger.comtag:blogger.com,1999:blog-319304714858376628.post-86362336130629519422010-12-07T15:15:33.695+01:002010-12-07T15:15:33.695+01:00Thanks for the link! I was not quite sure what the...Thanks for the link! I was not quite sure what the underlining was doing.<br /><br /> Let put absorb some of what you said. Thanks for the feedback.<br /><br />Cheers,<br /><br />Igor.Igorhttps://www.blogger.com/profile/17474880327699002140noreply@blogger.comtag:blogger.com,1999:blog-319304714858376628.post-76827851553047169112010-12-07T14:55:48.310+01:002010-12-07T14:55:48.310+01:00Unfortunately it looks like my link to "seemi...Unfortunately it looks like my link to "seemingly irrational approaches" has been screwed (or is this another anti-spam feature :-) ) so you didn't got the full context of my remark:<br /><br />http://econlog.econlib.org/archives/2010/11/the_park_ranger_2.html#124167<br /><br />You say:<br /><i>most times the practitioners of these models have their hands in the practical/experiential knowledge and are striving to include all their knowledge in these models.</i><br /><br />This is of course very commendable and obvious but the point about the (reciprocal, BTW) irreducibility between Métis and Techné is that, no matter what and how, the efficiency of Métis <i>cannot</i> be shoehorned into any kind of model because it always stays fuzzy.<br />This is why I say that it will eludes you, you are trying to recast a foreign method into your "Techné paradigm", this is known to be impossible since Aristotle.<br /><br />Yet, it <i>IS</i> possible to make good use of Métis, how do you think Roman engineers mostly worked given the paucity of their actual "hard knowledge"?<br /><br />Thus, no:<br /><br /><i>The people funding these models probably would not buy seemingly irrational approaches because, frankly, they are too many.</i><br /><br />There aren't "too many irrational approaches", this is what the work of Gerd Gigerenzer is about, evolution (surprise, surprise) has shaped up some heuristics which are actually effective under irreducible uncertainty but look profoundly irrational.<br /><br />What make me talk of "fascination" are, for instance the papers of Beauzamy about <i>Robust Mathematical Methods for Extremely Rare Events</i> or <i>The information associated with a [single] sample</i>, unfortunately this approach will <i>not</i> catch the valuable information under uncertainty you are longing for.Anonymousnoreply@blogger.comtag:blogger.com,1999:blog-319304714858376628.post-52653387149713461212010-12-07T10:45:50.448+01:002010-12-07T10:45:50.448+01:00I said as much for the "unknown unknowns"...I said as much for the "unknown unknowns" previously so no surprise there.<br /><br />However, I would beg to differ on the "known unknowns" that "refer to circumstances or outcomes that are known to be possible, but it is unknown whether or not they will be realized (no data or very low probability/extreme events)."<br /><br />Your point is that we are facing an irreducible difference between Metis and Techne ( Techne: abstract technical knowledge, Metis: practical, experiential knowledge) whereby one would make a difference between the modeler removed from the experience and the practitioner. So a bird's eye, dare I say philosophical, view of the issue would make it look like we can never know. But you'd be wrong, most times the practitioners of these models have their hands in the practical/experiential knowledge and are striving to include all their knowledge in these models. Deep down, the engineers working on the debris (columbia RMM#2) knew they were outside their database. The inability to convey that information at higher level of the hierarchy was the faulty part. The same thing happened to the Challenger. Somehow information degradation needed for hierarchical structures also seem at play in the "smoothing out" of low probability events. <br /><br />It really is not a question of fascination as in many instances of pure engineering like the building of a bridge or a nuclear power plant, you are *required* to have some grasp on what those extreme events/very low probability events are (As a matter of fact, most bridges these days follow guidelines set up by some of these early very low probability events). In order to do this, a substantial amount of funding is dedicated to this modeling. Just take a look at the RMM examples.<br /><br />The people funding these models probably would not buy seemingly irrational approaches because, frankly, they are too many. You are also presupposing that most engineering does not try these irrational approaches. Again, I can think of numerous examples where seemingly weird conditions are studied. Most casual observers would not know about it except for a few specialists. But eventually the issue is how do you constrain the cost of investigating a totally open set of conditions. <br /><br /><br />PS: I noted that moderation was not strong enough for some spam. I agree with you it sucks.Igorhttps://www.blogger.com/profile/17474880327699002140noreply@blogger.comtag:blogger.com,1999:blog-319304714858376628.post-71400442230754192912010-12-07T08:03:32.556+01:002010-12-07T08:03:32.556+01:00As well as your previous post about the solution o...As well as your previous post about the solution of "Selling from Novosibirsk" this is an instance of the irreducible difference between <i>Métis</i> and <i>Techné</i>.<br />The "unknown unknowns" (and even the known unknowns) are <i>outside</i> the reach of modeling and rationality this is why "rationalists" like you and Beauzamy are fascinated by these cases.<br />This will always eludes you, the best you can do is to accept <a rel="nofollow">seemingly irrational approaches</a> and swallow your pride.<br /><br />P.S. Enforcing registration for commenting while you are <i>also</i> moderating sucks and does not give you any extra guarantees about the content or authorship (another rationality failure ;-) )Anonymousnoreply@blogger.comtag:blogger.com,1999:blog-319304714858376628.post-30163202678955529772010-11-30T09:48:17.022+01:002010-11-30T09:48:17.022+01:00Yaroslav and Sergei,
If you read the comments (Y...Yaroslav and Sergei,<br /><br /><br />If you read the comments (Yaroslav I know you have), it looks like this community is slowly being made aware of the regularization issues but I think one of the issue the first commenter rightly points to is the need for them to agree that parcimony is a good thing or a recognizable fact. Then, they can go on and customize and improve the whole arsenal of tools used so far in Machine Learning, statistics, CS. Compressive sensing for instance is a nice framework in this context not because of the potential parcimony but rather because it is robust. This community is really looking at a vexing problem: they spend an enormous amount of resources in the description of the circuitry and so after all these efforts, modeling can only be a supplemental task to this endeavor. In other words, parcimony may not be an obvious interest to them considering the heavy investement of the description task.<br /><br />Igor.Igorhttps://www.blogger.com/profile/17474880327699002140noreply@blogger.comtag:blogger.com,1999:blog-319304714858376628.post-10254253030610902262010-11-28T09:37:31.765+01:002010-11-28T09:37:31.765+01:00I agree with Yaroslav. They use simple LSM without...I agree with Yaroslav. They use simple LSM without any regularizes or outliers rejection. No wonder they got nonsensical optimums. They should have tried L1 IRLS at least (my favorite :)Anonymousnoreply@blogger.comtag:blogger.com,1999:blog-319304714858376628.post-5082111421432533282010-11-27T18:31:58.958+01:002010-11-27T18:31:58.958+01:00They are defining optimal as the least squares fit...They are defining optimal as the least squares fit to the training data, adding that "conventional wisdom would indicate that the best parameter set is the one that minimizes the cost function, i.e. the best fit to the experimental data." Sounds like the concept of overfitting is not so widely known in computational biology.Yaroslav Bulatovhttps://www.blogger.com/profile/06139256691290554110noreply@blogger.comtag:blogger.com,1999:blog-319304714858376628.post-22206429787641625652010-11-27T17:16:29.538+01:002010-11-27T17:16:29.538+01:00Deconvolution of the point spread function of a co...Deconvolution of the point spread function of a computed tomography algorithm similarly leads to a solution of underdetermined equations that is far from any optimization algorithm. See:<br /><br />Dhawan, A.P., R. Gordon & R.M. Rangayyan (1984). Nevoscopy: three-dimensional computed tomography for nevi and melanomas in situ by transillumination. IEEE Trans. Med. Imaging MI-3(2), 54-61.<br /> <br />Dhawan, A.P., R.M. Rangayyan & R. Gordon (1984). Wiener filtering for deconvolution of geometric artifacts in limited-view image reconstruction. Proc. SPIE 515, 168-172.<br /> <br />Dhawan, A.P., R.M. Rangayyan & R. Gordon (1985). Image restoration by Wiener deconvolution in limited-view computed tomography. Applied Optics 24(23), 4013-4020.<br /> <br />Rangayyan, R.M., A.P. Dhawan & R. Gordon (1985). Algorithms for limited-view computed tomography: an annotated bibliography and a challenge. Applied Optics 24(23), 4000-4012.<br /> <br />Yours, -Dick Gordon gordonr@cc.umanitoba.caDick Gordonhttps://www.blogger.com/profile/15928227698528256439noreply@blogger.com